The models
Context and
cost, side
by side.
The two numbers that decide most model choices, for every model worth considering. No benchmark scores. We don’t run our own evaluations, and we won’t launder someone else’s through a table here.
Prices are USD per million tokens.
Verified 2026-08-07.
25 models
| Model | Developer | Tier | Context | In / 1M | Out / 1M | Relative output cost |
|---|---|---|---|---|---|---|
| Claude Fable 5 | Anthropic | Frontier | 1M | $10 | $50 | |
| Claude Opus 5 | Anthropic | Frontier | 1M | $5 | $25 | |
| Claude Sonnet 5 | Anthropic | Workhorse | 1M | $3 | $15 | |
| Claude Opus 4.8 | Anthropic | Frontier | 1M | $5 | $25 | |
| Claude Sonnet 4.6 | Anthropic | Workhorse | 1M | $3 | $15 | |
| Claude Haiku 4.5 | Anthropic | Efficient | 200K | $1 | $5 | |
| GPT-5.6 Sol | OpenAI | Frontier | 1M | $5 | $30 | |
| GPT-5.6 Terra | OpenAI | Workhorse | 1M | $2 | $12 | |
| GPT-5.6 Luna | OpenAI | Efficient | 1M | $0.20 | $1.20 | |
| GPT-5.5 | OpenAI | Frontier | 1M | $5 | $30 | |
| GPT-5.4 | OpenAI | Workhorse | - | $2.50 | $15 | |
| Gemini 3.1 Pro | Frontier | 2M | $2 | $12 | ||
| Gemini 3.6 Flash | Workhorse | 1M | $1.50 | $7.50 | ||
| Grok 4.5 | xAI | Frontier | - | $2 | - | not published |
| Grok 4.20 | xAI | Workhorse | - | $1.25 | $2.50 | |
| DeepSeek V4 ProOpen | DeepSeek | Efficient | 1M | $0.43 | $0.87 | |
| DeepSeek V4 FlashOpen | DeepSeek | Efficient | 1M | $0.14 | $0.28 | |
| Qwen 3.7 Max | Alibaba | Frontier | - | - | - | not published |
| Qwen 3.7 Flash | Alibaba | Efficient | - | $0.03 | $0.13 | |
| Qwen 3.6 27BOpen | Alibaba | Efficient | - | - | - | not published |
| Llama 4 ScoutOpen | Meta | Workhorse | 10M | - | - | not published |
| Llama 3.3 8BOpen | Meta | Efficient | 128K | - | - | not published |
| Mistral Large | Mistral AI | Workhorse | 128K | - | - | not published |
| GLM-5.2Open | Zhipu AI | Workhorse | - | - | - | not published |
| Kimi K2.6Open | Moonshot AI | Workhorse | - | - | - | not published |
A note on these numbers
An em dash means the figure is not published, or we could not source it well enough to print it. Several models use tiered pricing that rises above a context threshold, where we know about it, the model page says so. Provider documentation always wins over this table.
Full methodology →