Amazon (Amazon Nova & Amazon Titan on Amazon Bedrock)

AWS also sells Nova Forge (a build-your-own-frontier-model service) and Nova Act (an agent SDK), neither of which is a token-priced Bedrock model. The older Titan family is being wound down: only the Titan embedding models and Titan Image Generator G1 v2 still have Bedrock model cards, and the Titan Text generation models (Premier/Express/Lite) have been removed from the catalog.

A Access and limits

Access is via an AWS account -> Amazon Bedrock -> Model access, then the bedrock-runtime Converse/InvokeModel APIs signed with SigV4 (no standalone API key; Bedrock API keys are short- or long-term AWS credentials issued in the console). There is no free tier for Nova or Titan inference. IMPORTANT PRICING TIERS: (1) Routing tier - Nova 2 models have a cheaper "Global cross-region inference" price (global.* inference profile) and a ~10% higher "Geo cross-region / in-region" price (us.*/eu.*/jp.* profiles).

Prices recorded here are the STANDARD service tier at the cheapest published routing: Global for Nova 2 Lite/Pro/Omni, Geo/in-region for Nova Micro/Lite/Pro/Premier (which have no Global profile). Geo/in-region standard rates are Nova 2 Lite $0.33/$2.75, Nova 2 Pro $1.375/$11.00, Nova 2 Omni $0.30 text-in/$2.80 text-out. (2) Service tier - Standard (default), Priority = 1.75x standard (Nova 2 Lite $0.525/$4.375 global; Nova Pro $1.40/$5.60; Nova Premier $4.375/$21.875), Flex = 50% off standard (Nova 2 Lite $0.15/$1.25), Reserved (account-level commitment).

(3) Batch inference is 50% off standard. (4) Cache reads are 75% cheaper than the on-demand input price on Nova understanding models; explicit caching needs >=1K tokens per checkpoint, max 4 checkpoints, 5-minute TTL, max ~20K cached tokens. Nova Pro also has a latency-optimized option at $1.00 in / $4.00 out per 1M.

Built-in Web Grounding costs $30 per 1,000 requests on Nova 2 Lite, Nova 2 Pro, Nova 2 Omni and Nova Premier. Provisioned Throughput for Nova is $60.50/hr per model unit (no commitment), $55.00/hr (1-month), $30.25/hr (6-month); custom-model storage is $1.95/model/month. Fine-tuning (SFT) is $0.001/1K tokens (Micro), $0.002/1K (Lite), $0.008/1K (Pro), $0.00378/1K (Nova 2 Lite); RFT on Nova 2 Lite is $80/hour.

Regions vary sharply: Nova 2 Lite is broad (US, EU, JP, me-central-1), Nova Premier is us-east-1/us-east-2/us-west-2 only, Nova Multimodal Embeddings is us-east-1 + GovCloud, and Nova 2 Pro/Omni preview access requires an AWS account team request (Nova Forge customers get early access). Retired and excluded here: Titan Text Premier, Titan Text Express, Titan Text Lite and Titan Image Generator v1 - they still appear on the pricing page but no longer have model cards or regional-availability entries.

D Official references

• Homepage
• API documentation
• Pricing