§6

AI coding assistants and agents

This category covers tools that write, edit, run and review code on a developer's behalf — from terminal agents and IDE forks to browser app-builders and cloud agents that open pull requests unattended. As of September 2026 the market has consolidated hard: Cognition folded Windsurf into Devin Desktop, Cursor acquired Continue and was itself acquired by SpaceX, Roo Code's extension shut down and was archived, and Aider has not shipped a release in over a year.

Data checked 2026-09-06

Pricing has also converged on one shape — a flat subscription that buys a dollar-denominated credit or token allowance, after which you pay near-API rates — which means the sticker price is a floor, not a budget. The practical distinctions now are surface (terminal vs IDE vs cloud vs browser), whether you control model choice, and what an admin can actually restrict. Verify pricing before committing: every vendor in this list changed its plan structure at least once in the last twelve months.

A How to choose

Start with surface, because it is the only choice that is expensive to reverse. If your work lives in an existing repo on your machine, pick a terminal agent (Claude Code, Codex CLI, Copilot CLI, Cline) — they compose with git, CI and shell tooling, and you can swap them out in a day. If you want inline completion and tab-editing as well as agentic work, you need an IDE product (Cursor, Zed, Antigravity, Kiro), and you are accepting a fork or a new editor.

If the deliverable is a working prototype for a non-engineer, the browser builders (Lovable, v0, Bolt, Replit) beat any coding agent on time-to-first-URL and lose badly the moment the code has to move into your own infrastructure. Second axis: model control. Cline, Continue's open source, Aider and Qodo let you bring your own key and audit exactly what leaves your network; Amp and the browser builders pick models for you; Antigravity is Gemini-only.

Third axis: how the money works. A $20 plan that includes "$20 of usage" (Augment, Zed Pro, v0 Plus, Replit Core) is pass-through billing with a floor, so model the second, third and fourth hundred dollars, not the first twenty — Anthropic's own docs put enterprise Claude Code at roughly $13 per developer per active day and $150–250 per developer per month, which is the honest order of magnitude for any serious agentic use. Fourth axis: who administers it.

If you need SSO, SCIM, audit logs, spend caps and a defensible data-retention story today, the short list is Claude Code (Team/Enterprise, with Bedrock/Vertex/Foundry deployment), Copilot Business/Enterprise, and Cursor Enterprise; most of the rest sell "contact us." Two anti-defaults worth stating plainly: do not pick Cursor reflexively if your organisation has procurement concerns about its new SpaceX ownership or if you already pay for GitHub — Copilot at $19/seat with Copilot CLI covers a lot of the same ground — and do not pick a browser builder for anything you intend to maintain for more than a quarter. Finally, avoid anything on this list marked deprecated or discontinued (Windsurf, Roo Code, Continue's hosted product, arguably Aider) for new work, however good it was a year ago.

B At a glance

Name	Surface	Models	Terminal access	Price	Enterprise controls	Pricing
Claude Code	Terminal CLI, VS Code/Cursor extension, JetBrains plugin, desktop app (macOS/Windows, Linux beta), web + mobile, GitHub Actions, GitLab CI, Slack, Agent SDK	Claude family (Opus, Sonnet 4.6, Haiku, Fable); deployable via Anthropic API, Amazon Bedrock, Google Cloud Agent Platform, Microsoft Foundry or an LLM gateway	Yes — the CLI is the primary surface; pipes, -p non-interactive mode, hooks and shell tool use	Pro $20/mo; Max from $100/mo; Team $25/user/mo standard, $125/user/mo premium; Enterprise custom; or per-token API billing	Managed settings via MDM/managed-settings.json, org spend limits and spend reports, usage-credit toggles, Enterprise Analytics API, OpenTelemetry export, contracted-rate reporting, self-hosted gateway with per-user spend caps	Included with Claude Pro $20/mo ($17/mo billed annually), Max plans from $100/mo, Team seats $25/user/mo standard and $125/user/mo premium (monthly billing; $20 and $100 annual), Enterprise custom. API-key usage is billed per token instead. Anthropic's own docs cite ~$13 per developer per active day and $150-250 per developer per month across enterprise deployments.
OpenAI Codex	CLI (npm/Homebrew/install script), IDE extension, desktop app (Windows, macOS), web/cloud at chatgpt.com, SDK	GPT-6 Astra plus the GPT-5.6 family (Sol, Terra, Luna tiers); OpenAI models only	Yes — Codex CLI runs locally with sandboxing and approval modes	Free $0; Go $8/mo; Plus $20/mo; Pro from $100/mo; Business $20/user/mo; Enterprise custom; overflow via credits	ChatGPT Business/Enterprise workspace admin, pooled/flexible credits; detailed agent-specific policy controls not publicly documented	Included with ChatGPT plans: Free $0 (light use only), Go $8/mo, Plus $20/mo, Pro from $100/mo (stated as 5x or 20x Plus Codex usage), Business $20/user/mo, Enterprise custom. Overflow is bought as credits priced per model — e.g. GPT-6 Astra at 250 credits per 1M input tokens versus GPT-5.6 Luna at 5 credits per 1M input tokens.
Cursor	Desktop IDE (VS Code fork), Cursor CLI, cloud/background agents, team marketplace for rules, skills and plugins	Cursor Composer 2.5 and Grok 4.6/4.5 (co-trained with SpaceX AI), plus OpenAI GPT-5 series and Codex, Anthropic Claude, Google Gemini, Moonshot Kimi, Z.ai GLM 5.2	Yes — Cursor CLI plus the editor's integrated terminal and agent shell commands	Free; Pro $20/mo; Pro Plus $60/mo; Ultra $200/mo; Teams $40/user/mo; Enterprise custom	SAML/OIDC SSO, SCIM seat management, audit logs, service accounts, repository/model/MCP access controls, pooled usage, privacy mode, invoice/PO billing, AIUC-1 agent-security certification	Hobby free; Pro $20/mo; Pro Plus $60/mo; Ultra $200/mo; Teams $40/user/mo (billed per active user); Enterprise custom. A ₹649/mo India-only 'Start' plan covers the Cursor-models pool only. Model rates published per million tokens, e.g. Composer 2.5 $0.50 in / $2.50 out, Grok 4.6 $2 in / $6 out.
GitHub Copilot	VS Code, Visual Studio, JetBrains, Neovim, Eclipse, github.com, GitHub Mobile, Copilot CLI, cloud coding agent, PR code review	23+ options across Anthropic (Claude Opus, Haiku 4.5), OpenAI (GPT-5.6 variants, GPT-5 mini) and Google (Gemini Flash); admin-selectable	Yes — Copilot CLI is GA and included in all plans, with /plan, /fleet subagents and AGENTS.md support on macOS, Linux and Windows	Free $0; Pro $10/mo; Pro+ $39/mo; Max $100/mo; Business $19/user/mo; Enterprise $39/user/mo	Org policy management for models and preview features, audit logs, IP indemnity, content exclusion, budget/spend caps, SSO, per-org agent enablement	Free $0 (2,000 completions/mo, 50 chat requests); Pro $10/mo with $15 AI credits; Pro+ $39/mo with $70 credits; Max $100/mo with $200 credits; Copilot Business $19/user/mo; Copilot Enterprise $39/user/mo.
Devin	Web app (app.devin.ai) with embedded IDE, Slack, Microsoft Teams, Devin for Terminal CLI, API, Devin Desktop	Cognition SWE-1.7 plus third-party frontier models, routed automatically by Cognition's 'Adaptive' model router	Yes — Devin for Terminal CLI, and Devin has a full shell inside its own cloud workspace	Free; Pro $20/mo; Max $200/mo; Teams from $80/mo; Enterprise custom; on-demand credits beyond quota	Per-user and team ACU/credit caps via usage policies, group mappings, SSO, ACU consumption API, enterprise contract billing, federal deployment options	Self-serve plans: Free; Pro $20/mo; Max $200/mo; Teams from an $80/mo minimum; Enterprise on contract. Included daily/weekly usage quotas plus on-demand credits for overflow; legacy ACU plans are sunset, with credits described as the same dollar value as ACUs.
Windsurf (now Devin Desktop)	Devin Desktop IDE (successor); Windsurf Plugins for JetBrains and VS Code; legacy Windsurf enterprise auth	Cognition SWE-1.7 with the 'Adaptive' router over third-party frontier models	Yes — integrated terminal in Devin Desktop plus Devin for Terminal CLI	Folded into Devin plans: Free, Pro $20/mo, Max $200/mo, Teams from $80/mo, Enterprise custom	Devin's enterprise controls plus a legacy Windsurf enterprise authentication path for existing accounts	No standalone Windsurf pricing remains; access follows Devin's self-serve plans (Free, Pro $20/mo, Max $200/mo, Teams from $80/mo) or an enterprise contract.
Cline	VS Code extension, JetBrains plugin (IntelliJ, PyCharm, WebStorm, GoLand), CLI, web Kanban multi-agent board, Node.js SDK	BYO keys: Anthropic, OpenAI, Google, OpenRouter, AWS Bedrock, Azure/GCP Vertex, Cerebras, Groq, Ollama, LM Studio, any OpenAI-compatible endpoint; or Cline provider credits	Yes — a terminal CLI plus shell execution from the extension, gated by plan/act approval	Free software; pay-per-token inference (BYOK or Cline credits); Enterprise custom	Enterprise tier only: SSO/OIDC, role-based access control, centralised billing, team dashboards, audit logging, SLA and dedicated support	Extension, CLI and SDK are free (Apache 2.0). You pay only for inference — either your own provider keys at their published rates, or prepaid Cline provider credits. Enterprise pricing is custom (contact sales).
Roo Code	Was a VS Code extension; archived 15 May 2026. Successor Roomote is cloud or self-hosted with Slack, Teams, Discord, Telegram and web UI	Was BYO key across major providers (Cline-derived)	Was shell execution inside VS Code; no standalone CLI	Discontinued; Roomote cloud $49/$249/$499 per month by user band, self-hosted free up to 10 users	None current; Roomote offers self-hosting inside your own infrastructure as its main control story	No longer sold. Former paying users are directed to billing@roocode.com. The successor product Roomote is $49/mo (up to 10 users), $249/mo (11-50), $499/mo (51-100) on cloud; self-hosted is free up to 10 users.
Aider	Terminal only, with tight git integration (auto-commit per change)	BYO key across most providers and local models; shipped model list is stale (last updated for GPT-5 and Grok-4 in Aug 2025)	Yes — terminal is the only surface	$0 for the tool; provider API rates only	None	Free — install via pip. You pay only your LLM provider's API rates; no subscription, no seat fee, no vendor credits.
Sourcegraph Amp	Web (ampcode.com), macOS app, iOS app, CLI/terminal; persistent per-thread cloud 'orbs'; Puck agent coordinator	Chosen automatically by Amp — currently GPT-5.6, Claude Fable 5.1 and faster models by task; BYO inference keys on the Unconstrained plan	Yes — CLI runs locally or from your editor's terminal; orbs also provide a remote shell	Megawatt $20/mo; Gigawatt $200/mo; Unconstrained pay-as-you-go; education $10/mo	SSO, directory sync, admin controls, minimal data retention options; pay-as-you-go billing with BYO keys	Megawatt $20/mo (750 orb hours, $20 included agent usage, low/medium modes plus high with a linked ChatGPT account); Gigawatt $200/mo (1,000 xxlarge orb hours, $200 included agent usage, all modes including ultra); Unconstrained is pay-as-you-go at API rates for teams/enterprise with BYO inference keys. Students and teachers $10/mo. No free tier.
Zed	Native desktop editor with agent panel and edit prediction; supports external agents (e.g. Claude Code, Codex) inside the editor	Zed-hosted models on Pro/Business, or bring your own API keys on any plan; external agents run their own models	Yes — integrated terminal, and external CLI agents run inside it	Personal $0; Pro $10/mo (includes $5 of tokens); Business $30/seat/mo	Business tier: org-wide AI model policies and data governance over Zed-hosted services; no published SCIM or audit-log detail	Personal $0/mo (2,000 accepted edit predictions, unlimited use with your own API keys or external agents); Pro $10/mo (unlimited edit predictions, $5 of included tokens, usage-based billing after that); Business $30/seat/mo (unlimited predictions, no fixed AI credit allotment, hosted models at standard rates or BYO keys).
Continue	VS Code and JetBrains extensions, `cn` CLI, Continue Hub for shared configs and rules	Fully BYO — separate model roles for chat, autocomplete, edit and embeddings across any provider, including local	Yes — Continue CLI (`cn`)	unknown (pricing page removed after the Cursor acquisition); open source is free	unknown — previously offered via the hosted Hub, status uncertain post-acquisition	unknown — the pricing page no longer resolves following the Cursor acquisition. The open-source extensions and CLI are free and you supply your own model keys.
Qodo	Git provider PR reviews (GitHub, GitLab, Bitbucket), IDE integration for pre-PR review, CLI, analytics dashboard	Bring your own — OpenAI, Anthropic, Azure OpenAI or self-hosted models; default model not publicly stated	Partial — a CLI exists, but the product's centre of gravity is Git and IDE review, not a terminal agent	Free 14-day trial; Pro Team $30/mo up to 30 users with credit packs (2,500 / 5,000 / 20,000); Enterprise custom	SSO/SAML, audit logs, BYOK, single-tenant SaaS or on-premises deployment, advanced self-learning rules, dedicated CSM	14-day free trial with unlimited credits, no card. Pro Team $30/mo for up to 30 users, with credit packs sized at 2,500 credits (~18 reviews/mo), 5,000 (~36) and 20,000 (~144); monthly, no annual commitment. Enterprise is custom.
Augment Code (Cosmos)	Cosmos platform for SDLC agents (PR review, ticket-to-PR, vulnerability patching, incident response), CLI, MCP and native tools	unknown — Augment's Context Engine plus undisclosed frontier models; no public model list	Yes — CLI access is included on all plans	Standard $20/mo flat (includes $20 usage, up to 50 seats); Business $100/mo flat (includes $100 usage); Enterprise custom	SOC 2 Type II and no-training guarantee on all paid plans; SSO/OIDC/SCIM, CMEK, ISO 42001, multi-region compute, unlimited concurrent sessions and custom compute sizing on Enterprise only	Standard $20/mo flat, includes $20 of usage across LLM, context engine and compute, up to 50 seats. Business $100/mo flat, includes $100 of usage, up to 50 seats. Both support pay-as-you-go top-ups. Enterprise custom with unlimited users.
Replit Agent	Browser IDE and mobile app, with integrated hosting, database, auth and deployments	Undisclosed mix of third-party frontier models; credits are consumed at model-dependent rates	Yes — a shell inside the hosted workspace, but no agent on your local machine	Starter $0; Core $20/mo ($17 annual) with $20 credits; Pro $100/mo ($95 annual) with $100 credits; Enterprise custom	SSO/SAML, advanced privacy controls, single-tenant environments, static outbound IPs, custom seat limits — Enterprise only	Starter $0 (daily agent credits, 1 published project, one background task at a time); Core $20/mo, or $17/mo billed annually, with $20 in model credits; Pro $100/mo, or $95/mo billed annually, with $100 in credits, 10 parallel agents and up to 15 collaborators; Enterprise custom.
Lovable	Browser only, with Lovable Cloud hosting, database and auth; unlimited workspace members on all plans	Undisclosed frontier models; usage is abstracted entirely into credits	No	Free (5 credits/day); Pro from $25/mo for 100 credits to $2,250/mo for 10,000; Business from $50/mo to $4,300/mo; Enterprise volume contract	Enterprise plan with volume-based credit contracts; detailed SSO/audit-log specifics not published	Free: 5 daily build credits (max 30/mo) plus 20 monthly Cloud credits. Pro: credit tiers from 100 credits at $25/mo ($21 annual) up to 10,000 credits at $2,250/mo ($1,875 annual). Business: same tiers at roughly double, from 100 credits at $50/mo up to 10,000 at $4,300/mo. Enterprise: volume-based contract. All paid plans also include 5 daily build credits.
v0	Web app (v0.app) with chat, visual editor, GitHub sync and deploy to Vercel	Vercel's v0 models plus third-party frontier models routed through Vercel AI Gateway	No	Free $0 ($5 credits); Plus $30/user/mo ($30 credits + $2/day bonus); Business $100/user/mo; Enterprise custom	Business and Enterprise tiers with team seats and centralised billing; SSO and audit-log specifics are on the Vercel platform rather than published per-plan for v0	Free $0/mo with $5 of monthly credits; Plus $30/user/mo with $30 of monthly credits plus a $2 daily login bonus; Business $100/user/mo with $30 of monthly credits per user plus the daily bonus; Enterprise custom.
Bolt	Browser only, running the full toolchain in StackBlitz WebContainers; integrated hosting, database and auth	Standard and Max Pro model tiers at different token costs; specific underlying models not published	In-browser terminal via WebContainers; no access to a real OS shell or native binaries	Free $0 (300K/day, 1M/mo); Pro $25/mo from 10M tokens; Teams $30/member/mo; Enterprise custom; up to 28% off annually	Enterprise tier: SSO, audit logs, compliance support, 24/7 priority support	Free $0 (300K tokens/day, 1M tokens/month); Pro $25/mo starting at 10M tokens/month, no daily cap, unused tokens roll over; Teams $30/member/mo with the same 10M monthly tokens per member (tokens are not pooled across the team); Enterprise custom. Annual billing saves up to 28%.
Google Antigravity	Antigravity IDE, Antigravity CLI, Antigravity 2.0 multi-agent command centre, Python SDK; macOS (Apple Silicon and Intel) and other desktop builds	Google Gemini models including the Gemini 3.8 / 3.7 Flash line; no third-party models	Yes — Antigravity CLI is a terminal-first surface for the same agents	Free for individual developers; enterprise option with unpublished pricing	unknown — an enterprise offering exists but admin controls are not publicly documented	Available at no charge for individual developers, per the official site; an enterprise option exists with pricing not published. Underlying quota and rate limits are not stated publicly.
Kiro	Kiro IDE, CLI and web; spec-driven workflow producing requirements and design artefacts before implementation	Multiple models with premium models on paid plans; specific model list and versions not stated on the pricing page	Yes — CLI is a first-class surface alongside the IDE	Free $0 (50 credits); Pro $20/user/mo (1,000); Pro+ $40 (2,000); Pro Max $100 (5,000); Power $200 (10,000); extra credits $0.04 each	AWS account and Builder ID identity integration; per-plan admin, SSO and audit-log detail not published on the pricing page	Free $0 (50 credits/mo); Pro $20/user/mo (1,000 credits); Pro+ $40/user/mo (2,000); Pro Max $100/user/mo (5,000); Power $200/user/mo (10,000). Additional credits $0.04 each. New users get a $20 sign-up bonus when upgrading via social login or AWS Builder ID.

C Entries

Claude Code

A CLI agent that reads a repo, edits files, runs commands and opens PRs, with the same engine behind IDE extensions, a macOS/Windows desktop app, claude.ai/code in the browser, GitHub Actions and GitLab CI, Slack, and an Agent SDK. Configuration is file-based: CLAUDE.md for standing instructions, skills for packaged workflows, hooks to run shell commands around tool calls, and MCP for external systems. Unlike Cursor it is not an editor and has no tab-completion; unlike Devin it runs locally against your working tree by default rather than in a vendor sandbox. Enterprises can route it through Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or a self-hosted gateway instead of Anthropic's API.

Surface	Terminal CLI, VS Code/Cursor extension, JetBrains plugin, desktop app (macOS/Windows, Linux beta), web + mobile, GitHub Actions, GitLab CI, Slack, Agent SDK
Models	Claude family (Opus, Sonnet 4.6, Haiku, Fable); deployable via Anthropic API, Amazon Bedrock, Google Cloud Agent Platform, Microsoft Foundry or an LLM gateway
Terminal access	Yes — the CLI is the primary surface; pipes, -p non-interactive mode, hooks and shell tool use
Price	Pro $20/mo; Max from $100/mo; Team $25/user/mo standard, $125/user/mo premium; Enterprise custom; or per-token API billing
Enterprise controls	Managed settings via MDM/managed-settings.json, org spend limits and spend reports, usage-credit toggles, Enterprise Analytics API, OpenTelemetry export, contracted-rate reporting, self-hosted gateway with per-user spend caps

Watch out: It is not an IDE — there is no inline autocomplete or tab-edit model, so you keep your existing editor alongside it. Subscription usage is metered on rolling 5-hour and weekly windows, so a heavy session can stop mid-task until the window resets or you enable usage credits. Cost scales with context, not with effort: a long-running session re-sends the whole conversation on every turn, and agent teams use roughly 7x the tokens of a normal session. On Bedrock/Vertex/Foundry, Anthropic's analytics dashboards do not cover your usage at all — per-user attribution requires OpenTelemetry export or an LLM gateway.

Included with Claude Pro $20/mo ($17/mo billed annually), Max plans from $100/mo, Team seats $25/user/mo standard and $125/user/mo premium (monthly billing; $20 and $100 annual), Enterprise custom. API-key usage is billed per token instead. Anthropic's own docs cite ~$13 per developer per active day and $150-250 per developer per month across enterprise deployments.

OpenAI Codex

Codex is a local-and-cloud coding agent: a Rust/npm CLI (@openai/codex) that runs on your machine with sandboxing and approval prompts, an IDE extension, Windows and macOS desktop apps, and a hosted surface on chatgpt.com for tasks you kick off and check later. Authentication is normally your ChatGPT plan rather than an API key, which is what makes it cheap relative to metered alternatives. Against Claude Code the trade is ecosystem: Codex is tied to OpenAI's model line (GPT-6 Astra at the top, the GPT-5.6 family below it) with no BYO-model story, but it inherits ChatGPT's plan economics and workspace admin. Documentation moved to learn.chatgpt.com during 2026 and the old openai.com/codex URLs no longer serve.

Surface	CLI (npm/Homebrew/install script), IDE extension, desktop app (Windows, macOS), web/cloud at chatgpt.com, SDK
Models	GPT-6 Astra plus the GPT-5.6 family (Sol, Terra, Luna tiers); OpenAI models only
Terminal access	Yes — Codex CLI runs locally with sandboxing and approval modes
Price	Free $0; Go $8/mo; Plus $20/mo; Pro from $100/mo; Business $20/user/mo; Enterprise custom; overflow via credits
Enterprise controls	ChatGPT Business/Enterprise workspace admin, pooled/flexible credits; detailed agent-specific policy controls not publicly documented

Watch out: Message allowances are per five-hour window and vary by an order of magnitude across model tiers (roughly 5-45 messages on the top Astra tier versus 250-2,000 on the cheapest), so the top model is effectively rationed on Plus. There is no bring-your-own-model option and no non-OpenAI fallback. API-key authentication works but the docs describe it as requiring extra setup versus plan sign-in. Product URLs and doc structure have churned repeatedly in 2026, which makes older internal runbooks unreliable.

Included with ChatGPT plans: Free $0 (light use only), Go $8/mo, Plus $20/mo, Pro from $100/mo (stated as 5x or 20x Plus Codex usage), Business $20/user/mo, Enterprise custom. Overflow is bought as credits priced per model — e.g. GPT-6 Astra at 250 credits per 1M input tokens versus GPT-5.6 Luna at 5 credits per 1M input tokens. · open source

Cursor

Cursor is a desktop editor forked from VS Code that combines tab-completion, inline agent edits and background cloud agents, plus a CLI. It is unusual in shipping its own models — Composer 2.5 (listed at $0.50/$2.50 per million input/output tokens) and Grok 4.6/4.5, jointly trained with SpaceX's AI group — alongside GPT-5 series, Claude, Gemini, Kimi and GLM. Cursor acquired Continue in 2026 and was itself acquired by SpaceX on 14 August 2026. Paid plans buy a dollar-denominated usage pool rather than a request count, so your effective limit depends entirely on which model you select.

Surface	Desktop IDE (VS Code fork), Cursor CLI, cloud/background agents, team marketplace for rules, skills and plugins
Models	Cursor Composer 2.5 and Grok 4.6/4.5 (co-trained with SpaceX AI), plus OpenAI GPT-5 series and Codex, Anthropic Claude, Google Gemini, Moonshot Kimi, Z.ai GLM 5.2
Terminal access	Yes — Cursor CLI plus the editor's integrated terminal and agent shell commands
Price	Free; Pro $20/mo; Pro Plus $60/mo; Ultra $200/mo; Teams $40/user/mo; Enterprise custom
Enterprise controls	SAML/OIDC SSO, SCIM seat management, audit logs, service accounts, repository/model/MCP access controls, pooled usage, privacy mode, invoice/PO billing, AIUC-1 agent-security certification

Watch out: 'Extended limits' means a credit pool, not unlimited: choosing a frontier model instead of Composer drains a Pro plan quickly and spend is hard to forecast across a team. Being a VS Code fork means extension availability and upstream feature parity lag, and you are switching editors, not adding a tool. The SpaceX acquisition is a live procurement and data-governance question for regulated buyers who diligenced Anysphere. Teams billing is per active user, which is cheaper but makes month-to-month cost less predictable than fixed seats.

Hobby free; Pro $20/mo; Pro Plus $60/mo; Ultra $200/mo; Teams $40/user/mo (billed per active user); Enterprise custom. A ₹649/mo India-only 'Start' plan covers the Cursor-models pool only. Model rates published per million tokens, e.g. Composer 2.5 $0.50 in / $2.50 out, Grok 4.6 $2 in / $6 out.

GitHub Copilot

The broadest-surface product in the category: completions and chat in VS Code, Visual Studio, JetBrains, Neovim and Eclipse; a coding agent that works from GitHub issues and opens PRs; PR code review; and a generally available Copilot CLI with /plan, /fleet parallel subagents and per-task model switching, included in every plan including Free. Pricing moved from 'premium requests' to dollar-denominated AI credit allowances — $15 on Pro, $70 on Pro+, $200 on the new $100/mo Max tier. It is the default enterprise answer because it is already on the MSA, ships IP indemnity, and admins can restrict which of the 23-plus models are usable.

Surface	VS Code, Visual Studio, JetBrains, Neovim, Eclipse, github.com, GitHub Mobile, Copilot CLI, cloud coding agent, PR code review
Models	23+ options across Anthropic (Claude Opus, Haiku 4.5), OpenAI (GPT-5.6 variants, GPT-5 mini) and Google (Gemini Flash); admin-selectable
Terminal access	Yes — Copilot CLI is GA and included in all plans, with /plan, /fleet subagents and AGENTS.md support on macOS, Linux and Windows
Price	Free $0; Pro $10/mo; Pro+ $39/mo; Max $100/mo; Business $19/user/mo; Enterprise $39/user/mo
Enterprise controls	Org policy management for models and preview features, audit logs, IP indemnity, content exclusion, budget/spend caps, SSO, per-org agent enablement

Watch out: Business and Enterprise seat fees are on top of AI credit consumption, so the $19 or $39 line item is not the real cost of heavy agent use. The credit allowance runs out mid-month for anyone doing sustained agentic work on premium models, and admins who restrict models to control spend directly degrade agent quality. The Free tier's 50 chat requests per month is a trial, not a usable plan. Agent-mode quality has historically trailed Claude Code and Cursor on long multi-file tasks, even running the same underlying models.

Free $0 (2,000 completions/mo, 50 chat requests); Pro $10/mo with $15 AI credits; Pro+ $39/mo with $70 credits; Max $100/mo with $200 credits; Copilot Business $19/user/mo; Copilot Enterprise $39/user/mo.

Devin

Devin runs autonomously in its own cloud workspace — planning, editing, running tests and opening PRs — for tasks Cognition sizes at up to about three hours. You interact with it by tagging it in Slack or Microsoft Teams, from the web app at app.devin.ai, through Devin for Terminal, or via API. It is the opposite trade from Claude Code or Codex CLI: less interactive control, more unattended throughput, and work happens in Cognition's sandbox rather than your working tree. Cognition sunset the ACU billing metric during 2026 in favour of on-demand credits at the same dollar value, and now also owns the Windsurf lineage as Devin Desktop.

Surface	Web app (app.devin.ai) with embedded IDE, Slack, Microsoft Teams, Devin for Terminal CLI, API, Devin Desktop
Models	Cognition SWE-1.7 plus third-party frontier models, routed automatically by Cognition's 'Adaptive' model router
Terminal access	Yes — Devin for Terminal CLI, and Devin has a full shell inside its own cloud workspace
Price	Free; Pro $20/mo; Max $200/mo; Teams from $80/mo; Enterprise custom; on-demand credits beyond quota
Enterprise controls	Per-user and team ACU/credit caps via usage policies, group mappings, SSO, ACU consumption API, enterprise contract billing, federal deployment options

Watch out: Scoped to tasks of roughly three hours; ambiguous or architecturally novel work still needs a human driving. Autonomous runs are where credits disappear fastest, and a failed run still bills. Reliability depends heavily on repo onboarding — playbooks, knowledge and a working test command — so the first weeks are setup, not output. Cognition has restructured pricing at least twice (ACUs to credits), so any budget built from a 2025 quote is wrong.

Self-serve plans: Free; Pro $20/mo; Max $200/mo; Teams from an $80/mo minimum; Enterprise on contract. Included daily/weekly usage quotas plus on-demand credits for overflow; legacy ACU plans are sunset, with credits described as the same dollar value as ACUs.

Windsurf (now Devin Desktop)

Windsurf was the agentic IDE (formerly Codeium) acquired by Cognition on 14 July 2025. Through 2026 it was progressively merged with Devin's cloud agents — 'Devin in Windsurf' in April 2026 — and on 2 June 2026 Cognition shipped Devin Desktop, described as 'the next generation of Windsurf, built around Devin Cloud, the Agent Command Center, and a full IDE'. windsurf.com now 308-redirects to devin.ai/desktop and Cognition's docs carry a 'Legacy Windsurf Auth' page for existing enterprise accounts and a documented Windsurf-to-Devin-Desktop transition. Windsurf Plugins for JetBrains and VS Code still exist as separate integrations.

Surface	Devin Desktop IDE (successor); Windsurf Plugins for JetBrains and VS Code; legacy Windsurf enterprise auth
Models	Cognition SWE-1.7 with the 'Adaptive' router over third-party frontier models
Terminal access	Yes — integrated terminal in Devin Desktop plus Devin for Terminal CLI
Price	Folded into Devin plans: Free, Pro $20/mo, Max $200/mo, Teams from $80/mo, Enterprise custom
Enterprise controls	Devin's enterprise controls plus a legacy Windsurf enterprise authentication path for existing accounts

Watch out: Do not shortlist Windsurf as a current product: the brand, its own pricing page and its independent roadmap are gone. Existing customers face a migration to Devin Desktop with a legacy auth path, and any evaluation, benchmark or contract that names 'Windsurf' needs re-scoping. Model documentation for Devin Desktop is inconsistent — surviving pages still reference older model generations alongside SWE-1.7, so confirm what you actually get before signing.

No standalone Windsurf pricing remains; access follows Devin's self-serve plans (Free, Pro $20/mo, Max $200/mo, Teams from $80/mo) or an enterprise contract. · discontinued

Cline

Cline is the leading open-source coding agent: a VS Code extension, JetBrains plugin, CLI, Node SDK and a web Kanban board for running several agents at once, all under Apache 2.0. It uses plan/act modes with explicit human approval before edits and shell commands. The defining property is model neutrality — point it at Anthropic, OpenAI, Google, OpenRouter, Bedrock, Vertex, Cerebras, Groq, Ollama, LM Studio or anything OpenAI-compatible, or buy Cline's own inference credits. There is no seat fee for individuals; you pay for tokens.

Surface	VS Code extension, JetBrains plugin (IntelliJ, PyCharm, WebStorm, GoLand), CLI, web Kanban multi-agent board, Node.js SDK
Models	BYO keys: Anthropic, OpenAI, Google, OpenRouter, AWS Bedrock, Azure/GCP Vertex, Cerebras, Groq, Ollama, LM Studio, any OpenAI-compatible endpoint; or Cline provider credits
Terminal access	Yes — a terminal CLI plus shell execution from the extension, gated by plan/act approval
Price	Free software; pay-per-token inference (BYOK or Cline credits); Enterprise custom
Enterprise controls	Enterprise tier only: SSO/OIDC, role-based access control, centralised billing, team dashboards, audit logging, SLA and dedicated support

Watch out: No flat rate means no cost ceiling — a long agent run on a frontier model can cost more in an afternoon than a month of Cursor Pro, and you own the tuning (context trimming, model selection, cache behaviour) that bundled products do for you. There is no first-party model, so quality is entirely a function of whichever API you point it at. Enterprise features (SSO/OIDC, RBAC, audit logs, SLA) are behind an unpublished custom quote, which slows procurement compared with Copilot or Cursor.

Extension, CLI and SDK are free (Apache 2.0). You pay only for inference — either your own provider keys at their published rates, or prepaid Cline provider credits. Enterprise pricing is custom (contact sales). · open source

Roo Code

Roo Code was a popular Apache-2.0 fork of Cline that added multiple specialised agent modes inside VS Code. The extension was shut down on 15 May 2026 and the GitHub repository was archived read-only; the maintainers point users to ZooCode, a community fork, or back to Cline. The same team now ships Roomote, a separate source-available self-hosted cloud agent (Fair Core License with an Apache 2.0 future grant) that answers repo questions and opens PRs from Slack, Teams, Discord or Telegram, priced at $49/mo for up to 10 cloud users, $249/mo for 11-50 and $499/mo for 51-100, with self-hosting free up to 10 users.

Surface	Was a VS Code extension; archived 15 May 2026. Successor Roomote is cloud or self-hosted with Slack, Teams, Discord, Telegram and web UI
Models	Was BYO key across major providers (Cline-derived)
Terminal access	Was shell execution inside VS Code; no standalone CLI
Price	Discontinued; Roomote cloud $49/$249/$499 per month by user band, self-hosted free up to 10 users
Enterprise controls	None current; Roomote offers self-hosting inside your own infrastructure as its main control story

Watch out: Archived and unmaintained: no security patches, no new model support, and the extension itself was shut down rather than merely deprecated. ZooCode and other community forks carry no SLA and no guarantee of continuity. Roomote is a different product with a different shape (cloud/self-hosted agent, not an IDE extension) and a restrictive Fair Core License that caps non-commercial self-hosting at 10 users, so it is not a drop-in replacement.

No longer sold. Former paying users are directed to billing@roocode.com. The successor product Roomote is $49/mo (up to 10 users), $249/mo (11-50), $499/mo (51-100) on cloud; self-hosted is free up to 10 users. · open source · discontinued

Aider

Aider popularised the terminal coding-agent pattern: you chat in a shell, it applies edits as diffs and makes a git commit per change, so every step is revertable with normal git tooling. It works with almost any LLM through a provider key, including local models, and remains free and open source. The problem is momentum — the most recent release is v0.86.0 dated 9 August 2025, over a year old, and the homepage still advertises Claude 3.7 Sonnet and o1 as headline models. It is still a clean, dependency-light tool for scripted or offline use, but it is no longer a live project.

Surface	Terminal only, with tight git integration (auto-commit per change)
Models	BYO key across most providers and local models; shipped model list is stale (last updated for GPT-5 and Grok-4 in Aug 2025)
Terminal access	Yes — terminal is the only surface
Price	$0 for the tool; provider API rates only
Enterprise controls	None

Watch out: Effectively dormant: no release since 9 August 2025, so 2026 frontier models are not supported out of the box and the documentation references two-generation-old models. No MCP, no skills/hooks ecosystem, no sandboxing, no team or admin features, and no code review or CI integration. If you need an agent that will still be patched next quarter, choose Cline instead.

Free — install via pip. You pay only your LLM provider's API rates; no subscription, no seat fee, no vendor credits. · open source · unknown

Sourcegraph Amp

Amp runs each conversation thread on its own isolated cloud machine — an 'orb' — preconfigured with your code and tools, which continues executing after you disconnect; a component called Puck launches and coordinates multiple agents. It is available on the web, macOS and iOS apps, and a CLI you can run from any editor's terminal. Amp deliberately chooses models for you rather than exposing a picker, currently mixing GPT-5.6, Claude Fable 5.1 and cheaper models by task. Pricing is a subscription that includes a dollar amount of agent usage plus a quota of orb hours; there is no free tier at all.

Surface	Web (ampcode.com), macOS app, iOS app, CLI/terminal; persistent per-thread cloud 'orbs'; Puck agent coordinator
Models	Chosen automatically by Amp — currently GPT-5.6, Claude Fable 5.1 and faster models by task; BYO inference keys on the Unconstrained plan
Terminal access	Yes — CLI runs locally or from your editor's terminal; orbs also provide a remote shell
Price	Megawatt $20/mo; Gigawatt $200/mo; Unconstrained pay-as-you-go; education $10/mo
Enterprise controls	SSO, directory sync, admin controls, minimal data retention options; pay-as-you-go billing with BYO keys

Watch out: No free tier, so you cannot trial it without paying. The included agent usage equals the subscription price, meaning anything beyond light use bills at API rates on top — a heavy month on Megawatt costs multiples of $20. You do not choose models, which is a feature until a specific model is the one you need. Native apps are macOS and iOS only, so Windows and Linux users are limited to the CLI and web. Orb hours are a second meter to track alongside token spend.

Megawatt $20/mo (750 orb hours, $20 included agent usage, low/medium modes plus high with a linked ChatGPT account); Gigawatt $200/mo (1,000 xxlarge orb hours, $200 included agent usage, all modes including ultra); Unconstrained is pay-as-you-go at API rates for teams/enterprise with BYO inference keys. Students and teachers $10/mo. No free tier.

Zed

Zed is a GPU-accelerated editor written in Rust, built for low latency and multiplayer editing, with AI layered on: an edit-prediction model that completes as you type and an agent panel that can also drive external agents such as Claude Code or Codex through the agent client protocol. The editor itself is open source and free; the paid plans buy Zed-hosted inference and admin features. The Business tier at $30/seat/month is aimed at organisations that want model policy and data governance over Zed's hosted services while individual developers stay on their own API keys.

Surface	Native desktop editor with agent panel and edit prediction; supports external agents (e.g. Claude Code, Codex) inside the editor
Models	Zed-hosted models on Pro/Business, or bring your own API keys on any plan; external agents run their own models
Terminal access	Yes — integrated terminal, and external CLI agents run inside it
Price	Personal $0; Pro $10/mo (includes $5 of tokens); Business $30/seat/mo
Enterprise controls	Business tier: org-wide AI model policies and data governance over Zed-hosted services; no published SCIM or audit-log detail

Watch out: $5 of included tokens on Pro is negligible for agentic work — expect to be on usage billing or your own keys within days, which makes the $10 mostly a prediction-model subscription. The extension ecosystem is a fraction of VS Code's, so language and tooling support is uneven outside mainstream stacks. Business gives you governance over Zed's hosted services but the editor being open source means those controls do not cover everything a developer could do locally. Platform maturity is strongest on macOS and Linux.

Personal $0/mo (2,000 accepted edit predictions, unlimited use with your own API keys or external agents); Pro $10/mo (unlimited edit predictions, $5 of included tokens, usage-based billing after that); Business $30/seat/mo (unlimited predictions, no fixed AI credit allotment, hosted models at standard rates or BYO keys). · open source

Continue

Continue was the configuration-first alternative to closed assistants: an open-source VS Code and JetBrains extension plus a `cn` CLI where you declare which models handle chat, autocomplete, edit and embeddings, and a Hub for sharing those configurations, rules and MCP servers. In 2026 continue.dev was replaced by a 'Continue has joined Cursor' notice with FAQs covering existing subscriptions and data, and the pricing page no longer resolves; the team states the open-source codebase remains freely available. Treat the open source as forkable infrastructure and the hosted Hub and paid plans as winding down.

Surface	VS Code and JetBrains extensions, `cn` CLI, Continue Hub for shared configs and rules
Models	Fully BYO — separate model roles for chat, autocomplete, edit and embeddings across any provider, including local
Terminal access	Yes — Continue CLI (`cn`)
Price	unknown (pricing page removed after the Cursor acquisition); open source is free
Enterprise controls	unknown — previously offered via the hosted Hub, status uncertain post-acquisition

Watch out: The company is gone into Cursor, so the hosted Hub, paid plans and roadmap should be treated as end-of-life regardless of what the docs site still says; do not start a new team workflow on the Hub. Ongoing maintenance of the open source now depends on a community that has just lost its core maintainers. Anyone with a live subscription should confirm data handling and cancellation directly rather than relying on the docs, which carry no shutdown banner.

unknown — the pricing page no longer resolves following the Cursor acquisition. The open-source extensions and CLI are free and you supply your own model keys. · open source · deprecated

Qodo

Qodo has narrowed from a general assistant into a code-review and governance platform: agentic pull-request review with whole-codebase and cross-repository context, a 'living rules system' that encodes your standards and enforces them on every PR, pre-PR review in the IDE, and dashboards showing what was flagged and fixed. It is bring-your-own-model — OpenAI, Anthropic, Azure OpenAI or self-hosted — which matters for regulated buyers. Pricing is credit-based per review rather than per seat, so a 30-person team on the $30/mo Pro Team plan buys review volume, not licences.

Surface	Git provider PR reviews (GitHub, GitLab, Bitbucket), IDE integration for pre-PR review, CLI, analytics dashboard
Models	Bring your own — OpenAI, Anthropic, Azure OpenAI or self-hosted models; default model not publicly stated
Terminal access	Partial — a CLI exists, but the product's centre of gravity is Git and IDE review, not a terminal agent
Price	Free 14-day trial; Pro Team $30/mo up to 30 users with credit packs (2,500 / 5,000 / 20,000); Enterprise custom
Enterprise controls	SSO/SAML, audit logs, BYOK, single-tenant SaaS or on-premises deployment, advanced self-learning rules, dedicated CSM

Watch out: It is a reviewer, not a builder — the wrong choice if you want an agent that implements features in your editor. Credit-per-review pricing scales badly on high-PR-volume monorepos, where 144 reviews a month is a couple of days of traffic. The older Qodo Gen / Merge / Command product split has been consolidated into one platform, so tutorials and comparisons written in 2025 describe products that no longer exist under those names. Default model choice is not documented; the BYOK emphasis means quality depends on the key you supply.

14-day free trial with unlimited credits, no card. Pro Team $30/mo for up to 30 users, with credit packs sized at 2,500 credits (~18 reviews/mo), 5,000 (~36) and 20,000 (~144); monthly, no annual commitment. Enterprise is custom.

Augment Code (Cosmos)

Augment has repositioned from a context-heavy IDE assistant into Cosmos, a platform for running agents across the software lifecycle — automated PR review, ticket-to-PR, vulnerability patching and incident investigation — with a CLI, MCP and native tool access. Its pricing is the most unusual in the category: flat plans with no per-seat charge, up to 50 seats included, where the monthly fee also equals the included usage allowance ($20 buys $20 of LLM, context-engine and compute; $100 buys $100). Past that you top up pay-as-you-go. All paid tiers state no training on your code.

Surface	Cosmos platform for SDLC agents (PR review, ticket-to-PR, vulnerability patching, incident response), CLI, MCP and native tools
Models	unknown — Augment's Context Engine plus undisclosed frontier models; no public model list
Terminal access	Yes — CLI access is included on all plans
Price	Standard $20/mo flat (includes $20 usage, up to 50 seats); Business $100/mo flat (includes $100 usage); Enterprise custom
Enterprise controls	SOC 2 Type II and no-training guarantee on all paid plans; SSO/OIDC/SCIM, CMEK, ISO 42001, multi-region compute, unlimited concurrent sessions and custom compute sizing on Enterprise only

Watch out: Because the allowance equals the price, this is pass-through billing with a floor — the $20 plan is not a $20 answer for a team of 50 doing real work, it is $20 of tokens split 50 ways. The company has pivoted away from the retrieval-heavy IDE completion product many people know it for, so evaluations from 2025 do not describe today's offering. Model choice and the underlying providers are not publicly disclosed. CMEK, multi-region compute and SSO/OIDC/SCIM are Enterprise-only.

Standard $20/mo flat, includes $20 of usage across LLM, context engine and compute, up to 50 seats. Business $100/mo flat, includes $100 of usage, up to 50 seats. Both support pay-as-you-go top-ups. Enterprise custom with unlimited users.

Replit Agent

Replit Agent builds applications inside Replit's own hosted workspace and ships them on Replit's infrastructure — database, auth, hosting and domains included — so there is no local setup and no deploy step to wire up. Paid plans run parallel agents (10 on Pro) and background tasks, and credits are consumed against model cost. It differs from Lovable and v0 by being a real IDE underneath: you get a filesystem, a shell and package management, so you can drop out of the agent and edit code directly, which matters when the agent gets stuck.

Surface	Browser IDE and mobile app, with integrated hosting, database, auth and deployments
Models	Undisclosed mix of third-party frontier models; credits are consumed at model-dependent rates
Terminal access	Yes — a shell inside the hosted workspace, but no agent on your local machine
Price	Starter $0; Core $20/mo ($17 annual) with $20 credits; Pro $100/mo ($95 annual) with $100 credits; Enterprise custom
Enterprise controls	SSO/SAML, advanced privacy controls, single-tenant environments, static outbound IPs, custom seat limits — Enterprise only

Watch out: You are building inside Replit's hosting — moving a mature app to your own infrastructure means untangling their database, auth and deploy conventions. Credits are consumed by retries as well as successes, so a debugging spiral costs real money. It is a poor fit for large existing repositories or for code you are not permitted to upload. SSO/SAML, single-tenant environments and static outbound IPs are Enterprise-only.

Starter $0 (daily agent credits, 1 published project, one background task at a time); Core $20/mo, or $17/mo billed annually, with $20 in model credits; Pro $100/mo, or $95/mo billed annually, with $100 in credits, 10 parallel agents and up to 15 collaborators; Enterprise custom.

Lovable

Lovable turns a prompt into a deployed full-stack web app in the browser, with its own hosting and cloud services. Its pricing is unusually granular: everything is denominated in credits, plans are sold as credit tiers rather than feature tiers, and individual prompts cost fractional credits — the docs give examples from 0.50 credits for 'make the button gray' to 1.70 for 'build me a landing page, use images'. Workspaces allow unlimited members on every plan, so you buy throughput, not seats. Free users get 5 build credits a day capped at 30 a month.

Surface	Browser only, with Lovable Cloud hosting, database and auth; unlimited workspace members on all plans
Models	Undisclosed frontier models; usage is abstracted entirely into credits
Terminal access	No
Price	Free (5 credits/day); Pro from $25/mo for 100 credits to $2,250/mo for 10,000; Business from $50/mo to $4,300/mo; Enterprise volume contract
Enterprise controls	Enterprise plan with volume-based credit contracts; detailed SSO/audit-log specifics not published

Watch out: There is no terminal and no local development loop — you cannot drop into a shell when the agent goes wrong, which is the single biggest difference from Replit. Every debugging iteration costs credits, so a stubborn bug is billed repeatedly at fractions of a credit per prompt. It is the wrong tool for an existing codebase, for anything needing bespoke infrastructure, or for code that must pass a security review you control. Credit consumption per prompt is variable and hard to forecast, and the top Pro tier at $2,250/mo is real money for a hosted builder.

Free: 5 daily build credits (max 30/mo) plus 20 monthly Cloud credits. Pro: credit tiers from 100 credits at $25/mo ($21 annual) up to 10,000 credits at $2,250/mo ($1,875 annual). Business: same tiers at roughly double, from 100 credits at $50/mo up to 10,000 at $4,300/mo. Enterprise: volume-based contract. All paid plans also include 5 daily build credits.

v0

v0 generates and iterates full-stack web apps from a chat interface, with a visual editor for direct manipulation, GitHub sync so generated code lands in a real repo, and one-click deploy to Vercel. Its output is opinionated toward Next.js, React, Tailwind and shadcn/ui, which is exactly why it produces working, idiomatic code faster than general-purpose agents in that stack. Credits are dollar-denominated and, on Plus and Business, the monthly credit allowance equals or trails the subscription price, with a $2 daily login bonus on top.

Surface	Web app (v0.app) with chat, visual editor, GitHub sync and deploy to Vercel
Models	Vercel's v0 models plus third-party frontier models routed through Vercel AI Gateway
Terminal access	No
Price	Free $0 ($5 credits); Plus $30/user/mo ($30 credits + $2/day bonus); Business $100/user/mo; Enterprise custom
Enterprise controls	Business and Enterprise tiers with team seats and centralised billing; SSO and audit-log specifics are on the Vercel platform rather than published per-plan for v0

Watch out: Heavily biased to the Vercel stack — Next.js, React, Tailwind, shadcn/ui — and fighting that opinion is slow and expensive. Business costs $100/user/month but is documented as including the same $30 of monthly credits per user as Plus, so the extra is buying seats and controls, not capacity. There is no terminal and it is not a repo-wide agent: it generates and edits projects rather than working across a large existing codebase. Credits map to model spend, so complex generations burn the monthly allowance quickly.

Free $0/mo with $5 of monthly credits; Plus $30/user/mo with $30 of monthly credits plus a $2 daily login bonus; Business $100/user/mo with $30 of monthly credits per user plus the daily bonus; Enterprise custom.

Bolt

Bolt runs Node, npm and a dev server entirely inside the browser using StackBlitz's WebContainers, so the agent installs packages, runs the app and iterates without any cloud VM or local setup. Pricing is metered in raw tokens rather than abstract credits, which makes the cost model unusually legible: the free tier gives 300K tokens a day capped at 1M a month, Pro starts at 10M tokens a month with unused tokens rolling over and no daily cap. Beyond web apps it now markets slide decks, dashboards and storefronts, aimed as much at PMs and marketers as at developers.

Surface	Browser only, running the full toolchain in StackBlitz WebContainers; integrated hosting, database and auth
Models	Standard and Max Pro model tiers at different token costs; specific underlying models not published
Terminal access	In-browser terminal via WebContainers; no access to a real OS shell or native binaries
Price	Free $0 (300K/day, 1M/mo); Pro $25/mo from 10M tokens; Teams $30/member/mo; Enterprise custom; up to 28% off annually
Enterprise controls	Enterprise tier: SSO, audit logs, compliance support, 24/7 priority support

Watch out: Everything runs in WebContainers, so anything requiring native binaries, Docker, or a non-JavaScript runtime is out of scope. Token metering punishes long debugging sessions directly — a stuck agent re-reading a large project is expensive. Team tokens are explicitly not shared between members, so a heavy user cannot draw on a quiet colleague's allowance. The marketing now targets non-developers, and claims like '98% less errors' should be treated as vendor copy, not a benchmark.

Free $0 (300K tokens/day, 1M tokens/month); Pro $25/mo starting at 10M tokens/month, no daily cap, unused tokens roll over; Teams $30/member/mo with the same 10M monthly tokens per member (tokens are not pooled across the team); Enterprise custom. Annual billing saves up to 28%.

Google Antigravity

Antigravity is Google's agent-first development platform: an agentic IDE, a terminal-first CLI, an SDK for prototyping custom agents in Python, and Antigravity 2.0, a command centre for supervising many local agents running in parallel across projects. It runs Gemini models, including the Gemini 3.x Flash line, and is offered at no charge to individual developers with a separate enterprise option. The free-for-individuals position makes it the cheapest way to get a full agentic IDE plus CLI, at the cost of being locked to one model family.

Surface	Antigravity IDE, Antigravity CLI, Antigravity 2.0 multi-agent command centre, Python SDK; macOS (Apple Silicon and Intel) and other desktop builds
Models	Google Gemini models including the Gemini 3.8 / 3.7 Flash line; no third-party models
Terminal access	Yes — Antigravity CLI is a terminal-first surface for the same agents
Price	Free for individual developers; enterprise option with unpublished pricing
Enterprise controls	unknown — an enterprise offering exists but admin controls are not publicly documented

Watch out: Gemini-only — there is no bring-your-own-model path to Claude or GPT, so you cannot fall back when Gemini is the wrong tool for a task. 'Free' means quota-limited and subject to change rather than guaranteed capacity, and the free tier's limits are not published. The product line has been renamed and restructured since launch and is iterating fast, which means breaking changes and short-lived documentation. Enterprise terms and admin controls are not publicly documented, so procurement is a conversation, not a signup.

Available at no charge for individual developers, per the official site; an enterprise option exists with pricing not published. Underlying quota and rate limits are not stated publicly.

Kiro

Kiro is AWS's agentic development tool, distinguished by a spec-driven workflow: you produce requirements and design documents that the agent then implements against, rather than prompting straight into code. It runs in an IDE, a CLI and on the web, and bills in credits where a credit is 'a unit of work in response to user prompts', consumed faster by more complex prompts and more capable models. The tiering is unusually fine-grained — five paid steps from $20 to $200 a month — and extra credits are sold at a flat $0.04 each, which makes overflow cost easy to compute.

Surface	Kiro IDE, CLI and web; spec-driven workflow producing requirements and design artefacts before implementation
Models	Multiple models with premium models on paid plans; specific model list and versions not stated on the pricing page
Terminal access	Yes — CLI is a first-class surface alongside the IDE
Price	Free $0 (50 credits); Pro $20/user/mo (1,000); Pro+ $40 (2,000); Pro Max $100 (5,000); Power $200 (10,000); extra credits $0.04 each
Enterprise controls	AWS account and Builder ID identity integration; per-plan admin, SSO and audit-log detail not published on the pricing page

Watch out: A 'credit' is an opaque unit that varies with prompt complexity and model choice, so the headline credit counts do not translate cleanly into tasks completed — budget from the $0.04 overflow rate instead. Sign-up and org use route through AWS Builder ID or an AWS account, which is friction if you are not already an AWS shop. The spec-first workflow is genuine overhead on small changes where a terminal agent would just do the work. Ecosystem, extension support and model refresh cadence trail Cursor and Copilot.

Free $0 (50 credits/mo); Pro $20/user/mo (1,000 credits); Pro+ $40/user/mo (2,000); Pro Max $100/user/mo (5,000); Power $200/user/mo (10,000). Additional credits $0.04 each. New users get a $20 sign-up bonus when upgrading via social login or AWS Builder ID.