Cheapest capable models
Lowest blended cost per million tokens among generally-available models.
- Qwen3.7-Flash $0.06/M blended
- Amazon Nova Micro $0.06/M blended
- Command R7B $0.07/M blended
- Qwen-VL-OCR $0.09/M blended
- Ministral 3 3B $0.10/M blended
The AI services reference
Specifications, prices and trade-offs for 419 models from 13 providers — plus the platforms, tools and protocols around them. Collected from vendors' own documentation, dated, and linked back to source.
Data checked 2026-09-06§1 Flagship models at a glance
Blended cost assumes three input tokens per output token — a fairer single number than input price alone, which flatters models with cheap input and expensive output.
| Model | Provider | Context | Max out | In $/M | Out $/M | Blended | Status |
|---|---|---|---|---|---|---|---|
GPT-6 Astra
gpt-6-astra
|
OpenAI | 1.1M | 128K | $10.00 | $50.00 | $20.00 | ga |
GPT-5.5 Pro
gpt-5.5-pro
|
OpenAI | 1.1M | 128K | $30.00 | $180.00 | $67.50 | ga |
Gemini 3.8 Flash
gemini-3.8-flash
|
Google DeepMind | 1M | 65.5K | $0.75 | $3.75 | $1.50 | ga |
Muse Spark 1.3
muse-spark-1.3
|
Meta | 1M | — | $1.25 | $4.25 | $2.00 | ga |
Kimi K3
kimi-k3
|
Chinese AI labs | 1M | — | $2.98 | $14.90 | $5.96 | ga open |
Claude Fable 5.1
claude-fable-5-1
|
Anthropic | 1M | 128K | $10.00 | $50.00 | $20.00 | ga |
Grok 4.3
grok-4.3
|
xAI | 1M | — | $1.25 | $2.50 | $1.56 | ga |
DeepSeek-V4-Flash
deepseek-v4-flash
|
DeepSeek | 1M | 384K | $0.44 | $1.32 | $0.66 | ga open |
DeepSeek-V4-Pro
deepseek-v4-pro
|
DeepSeek | 1M | 384K | $1.32 | $3.96 | $1.98 | ga open |
Qwen3.8-Max
qwen3.8-max
|
Alibaba Qwen | 1M | 131.1K | $2.00 | $6.00 | $3.00 | ga |
Qwen3.7-Max
qwen3.7-max
|
Alibaba Qwen | 1M | 131.1K | $2.50 | $7.50 | $3.75 | ga |
Amazon Nova 2 Lite
amazon.nova-2-lite-v1:0
|
Amazon (Amazon Nova & Amazon Titan on Amazon Bedrock) | 1M | 65.5K | $0.30 | $2.50 | $0.85 | ga |
Widest context windows
Most input tokens accepted in a single request.
- Llama 4 Scout 10M tokens
- GPT-6 Astra 1.1M tokens
- GPT-5.6 Sol 1.1M tokens
- GPT-5.6 Terra 1.1M tokens
- GPT-5.6 Luna 1.1M tokens
§2 Start here
Model index
Every current model with context, pricing, modalities and capabilities. Sortable and filterable.
Cost calculator
Put your own token volumes in and see what each model would actually cost per month.
Compare models
Put up to four models side by side; differing rows are highlighted.
Providers
Who makes what, how to get an API key, and how each provider's pricing is structured.
Guides
How to choose a model, control cost, and read benchmarks without being misled.
The ecosystem
Image, video and audio models, inference platforms, vector databases, agent frameworks and more.
§3 Recent guides
How to choose a model
A task-shape-first method for picking a model tier: define the shape, set a capability floor and a cost-per-completed-task ceiling, benchmark on a private eval, then step down from frontier until quality breaks.
Understanding token pricing
How LLM API billing actually works: the four input meters, why resent conversation history dominates the bill, and why cost per completed task is the only number that ranks options correctly.
Context windows in practice
How token windows actually behave under load: why input and output limits are separate, why advertised length overstates useful length, what context costs, and which management strategy to reach for first.
How to read AI benchmarks
A working guide to benchmark families, contamination, saturation, harness effects, and the statistics that decide whether a leaderboard gap is real — plus a protocol for evaluating models on your own task.
§4 Browse the ecosystem
Image generation models
Text-to-image models are now bought like any other API primitive: metered per image or per megapixel, in the $0.01–$0.15 band, with editing and multi-reference composition as table stakes rather than separate products.
Video generation models
Text-to-video and image-to-video models turn a prompt, a still frame, or reference footage into a short clip — now routinely with a synchronized audio track generated in the same pass.
Speech, voice and music models
This category covers three related but commercially distinct markets: text-to-speech (TTS), speech-to-text (STT/ASR), and generative music.
Embedding and reranking models
Embedding models turn text (and increasingly images, audio and video) into dense vectors so that semantic similarity becomes a distance computation; rerankers are cross-encoders that take a query plus a candidate list and re-score them jointly, trading throughput for precision.
Hosted inference for open models
Hosted inference vendors run open-weight models on their own hardware and sell them as a metered API, so you get open-model economics and licence freedom without operating GPUs.
Cloud AI platforms
Cloud AI platforms are the managed layers that hyperscalers and data platforms put in front of foundation models: one API endpoint, one billing relationship, and the cloud's existing identity, networking and audit controls.
LLM gateways and routers
An LLM gateway sits between your application and one or more model providers, exposing a single OpenAI-compatible (and increasingly Anthropic-compatible) endpoint while handling key management, provider failover, retries, rate limits, caching, spend attribution and audit logs.
Vector databases and retrieval infrastructure
A vector database stores high-dimensional embeddings and answers approximate-nearest-neighbour (ANN) queries over them, almost always combined with metadata filters and increasingly with lexical BM25 scoring in the same query.
Agent and application frameworks
Agent frameworks sit between your application code and a model API, supplying the tool-calling loop, conversation state, streaming plumbing and multi-agent orchestration you would otherwise hand-write.
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.
LLM observability and evaluation
LLM observability and evaluation tools capture what an LLM app or agent actually did — the prompts, tool calls, retrieved documents, token counts, latency and cost of every span — and then let you score that behaviour, either offline against a dataset or online against live traffic.
Consumer AI assistants and AI search
This category covers the general-purpose chat assistants and answer engines that ordinary people and knowledge workers use directly, as opposed to APIs or coding agents.
Benchmarks and leaderboards
Every model launch cites benchmark numbers, and almost every one of those numbers is less informative than it looks.
Protocols and standards
The AI stack spent 2023-2024 with no interop layer at all: every tool integration, every agent handoff, every trace format was bespoke.
Document parsing, scraping and AI search APIs
This category covers the three services that sit between raw sources and an LLM's context window: document parsers that turn PDFs, scans and Office files into markdown or JSON; web scrapers that turn URLs into clean text while handling JavaScript and anti-bot defences; and search APIs that return ranked, LLM-ready results or grounded answers.
Fine-tuning and training platforms
Fine-tuning is how you get a smaller, cheaper, or more obedient model to beat a large general model on one narrow task — format adherence, domain jargon, classification, tool-call shape, latency and cost per request.
AI safety, guardrails and governance
This category covers the runtime controls that sit between users, LLMs and tools — classifiers that score prompts and completions for harm, prompt-injection and jailbreak detectors, PII redaction, groundedness/hallucination checks and topic restriction — plus the governance frameworks that auditors and regulators actually ask about.
Agent sandboxes and code execution
Agent frameworks produce code; something has to run it.
Self-hosted serving engines and inference servers
This is the software that turns a folder of open weights into an HTTP endpoint.
Managed RAG and enterprise search platforms
These are the products you buy instead of assembling a parser, an embedding model, a vector store, a reranker and a connector fleet yourself.
Voice agent orchestration and realtime infrastructure
A speech model list stops where the hard part starts.
Agent memory and context persistence
Agent memory tooling answers two questions a raw vector database does not: what is worth writing down from a conversation, and what small subset of that gets injected into the next prompt.
Data labeling, RLHF and human evaluation vendors
This category covers the companies that supply the human side of model training: annotated datasets, supervised fine-tuning demonstrations, preference and reward signal for RLHF, RL environments, red-teaming, and human evaluation panels.
AI red teaming and pre-deployment security testing
AI red teaming is the offensive counterpart to runtime guardrails: before an LLM application ships, someone deliberately attempts prompt injection, jailbreaks, tool abuse, authorisation bypass and data exfiltration against the real system, and writes down what worked.