What every major model actually costs per million tokens — GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and more. Sort any column, plug in your own workload, and compare models side by side.
| LINKS | |||||||
|---|---|---|---|---|---|---|---|
GPT-5.6 Sol OpenAI · flagship | $4.00 | $0.40 | $20.00 | 1.05M | $14.00 | CONSOLEOPENROUTER | |
GPT-5.6 Terra OpenAI | $2.00 | $0.20 | $12.00 | 1.05M | $8.00 | CONSOLEOPENROUTER | |
GPT-5.5 OpenAI · prev flagship | $5.00 | $0.50 | $30.00 | 1.05M | $20.00 | CONSOLEOPENROUTER | |
GPT-5.6 Luna OpenAI · fast tier | $0.20 | $0.02 | $1.20 | 1.05M | $0.80 | CONSOLEOPENROUTER | |
GPT-5.4 Mini OpenAI | $0.75 | $0.075 | $4.50 | 400K | $3.00 | CONSOLEOPENROUTER | |
GPT-5.4 Nano OpenAI | $0.20 | — | $1.25 | 400K | $0.82 | CONSOLEOPENROUTER | |
Claude Fable 5 Anthropic · top capability | $10.00 | $1.00 | $50.00 | 1M | $35.00 | CONSOLEOPENROUTER | |
Claude Opus 5 Anthropic · flagship | $5.00 | $0.50 | $25.00 | 1M | $17.50 | CONSOLEOPENROUTER | |
Claude Opus 4.8 Anthropic · prev flagship | $5.00 | $0.50 | $25.00 | 1M | $17.50 | CONSOLEOPENROUTER | |
Claude Sonnet 5 Anthropic | $2.00 | $0.20 | $10.00 | 1M | $7.00 | CONSOLEOPENROUTER | |
Claude Haiku 4.5 Anthropic · fast tier | $1.00 | $0.10 | $5.00 | 200K | $3.50 | CONSOLEOPENROUTER | |
Gemini 3.1 Pro Google · >200K: $4/$18 | $2.00 | $0.20 | $12.00 | 1.05M | $8.00 | CONSOLEOPENROUTER |
All prices in USD per 1M tokens, from official provider pricing pages as of August 30, 2026. Tiered long-context surcharges noted per model. Spot an outdated price? Tell us.
Every LLM API bill is the same formula: input tokens × input rate + output tokens × output rate. Rates are quoted per million tokens, and output typically costs 3–6× more than input. Long system prompts, retries, and verbose responses are where budgets quietly die.
Cached input is the biggest lever most teams ignore: resend the same prompt prefix and providers charge roughly a tenth of the fresh-input rate. If you run agents or RAG with a long, stable system prompt, cache pricing — not model choice — often decides your bill.
Prices moved fast this year, almost always downward. We verify every number on this page against the provider’s official pricing docs and stamp the date up top.
Every major provider bills per token, with separate rates for input (your prompt) and output (the model’s reply). Prices are quoted per million tokens. Output tokens usually cost 3–6× more than input tokens, so response length drives most of the bill for chat-style workloads.
A token is a chunk of text — roughly 4 characters or ¾ of an English word. “How much does the Claude API cost?” is about 9 tokens. A million tokens is roughly 750,000 words, or about ten novels.
If you resend the same prompt prefix (a system prompt, few-shot examples, long documents), providers can serve it from cache at a steep discount — typically 10× cheaper than fresh input tokens. Agents and RAG apps with long stable prefixes save the most.
Depends on the tier you need. Among frontier flagships, mid-tier models like Claude Sonnet 5 and GPT-5.6 Terra cost a fraction of the top models, and small-tier models (GPT-5.6 Luna, Gemini Flash-Lite, Haiku 4.5) are 10–50× cheaper still. Open-weight models served via API are often the lowest ¢/token — sort the table by output price and compare against your quality bar.
Prices are verified against each provider’s official pricing page — last check August 30, 2026. Vendors change prices often (usually downward, sometimes with promo windows). Spot something stale? Email support@reqkey.com and we’ll fix it within a day.
ReqKey gives every API key a credit balance — validate a key, deduct credits, enforce limits, and see spend per consumer in real time. One call from your own middleware.