Compare
The families at a glance, then a calculator. The calculator is deliberately simple: it ignores batching, caching discounts, autoscaling and throughput limits. Use it to see the shape of the trade, not to set a budget.
Side by side 8
| Family | Kind | Licence | Max context | Sizes / $ in–out per 1M | Access | Checked |
|---|---|---|---|---|---|---|
| Claude Anthropic | Frontier | Proprietary | 1M |
Claude Fable 5.1: $10 / $50
Claude Opus 5.5: $4 / $20
Claude Sonnet 5.5: $2 / $10
Claude Haiku 5.5: $0.1 / $0.5
Claude Opus 5: $5 / $25
|
APIBedrockVertex AIAzure | 2026-10-23 |
| GPT OpenAI | Frontier | Proprietary | 1.05M |
GPT-6 Astra: $10 / $50
GPT-6.1 Sol: $2 / $10
GPT-6 Luna: $0.1 / $0.5
GPT-5.6 Sol: $4 / $20
|
APIBedrockAzure | 2026-10-23 |
| Gemini Google | Frontier | Proprietary | 1.04858M |
Gemini 4 Argon: $2 / $10
Gemini 3.8 Flash: $0.75 / $3.75
Gemini 3.5 Flash-Lite: $0.3 / $2.5
Gemini 3.1 Pro (preview): $2 / $12
|
APIVertex AI | 2026-10-23 |
| Llama & Muse Meta | Open-weight | Open, restricted | 10M | 29.6B – 816B | HFBedrockSageMaker JumpStart | 2026-10-23 |
| Mistral Mistral AI | Open-weight | Open source | 262.144K | 3.9B – 675B | APIBedrockSageMaker JumpStartHF | 2026-10-23 |
| Qwen Alibaba | Open-weight | Open, restricted | 262.144K | 27B – 2400B | APIHF | 2026-10-23 |
| Gemma Google | Open-weight | Open source | 262.144K | 5.1B – 31.3B | HFBedrockSageMaker JumpStart | 2026-10-23 |
| DeepSeek DeepSeek | Open-weight | Open source | 1.04858M | 284B – 1600B | APIBedrockHF | 2026-10-23 |
Monthly cost
API cost = input tokens × input price + output tokens × output price. Self-host cost = one instance's hourly rate × hours running. Self-hosting is billed while the instance is up, busy or not; 730 hours is a full month.
| Model | How it is billed | Rate | Monthly cost |
|---|
Frontier prices are list prices from each provider, standard tier. Self-host rates are approximate on-demand SageMaker real-time inference prices, us-east-1, checked 2026-10-23, using the cheapest instance in the model's hardware tier at 16-bit precision; a "+" marks a floor where one instance is not enough. One instance's throughput is finite: at high volume you need more of them. Rows with "—" have an unverified price and are not costed.