Every current Claude model with its real per-million-token price — input, output, and cache reads — plus a calculator for your own workload and a straight comparison against GPT and Gemini.
| MODEL | INPUT $/1M | CACHED IN | OUTPUT $/1M | BLENDED | CONTEXT | YOUR COST |
|---|---|---|---|---|---|---|
Claude Fable 5 claude-fable-5 · top capability | $10.00 | $1.00 | $50.00 | $20.00 | 1M | $35.00 |
Claude Opus 5 claude-opus-5 · flagship | $5.00 | $0.50 | $25.00 | $10.00 | 1M | $17.50 |
Claude Opus 4.8 claude-opus-4-8 · prev flagship | $5.00 | $0.50 | $25.00 | $10.00 | 1M | $17.50 |
Claude Sonnet 5 claude-sonnet-5 | $2.00 | $0.20 | $10.00 | $4.00 | 1M | $7.00 |
Claude Haiku 4.5 claude-haiku-4-5 · fast tier | $1.00 | $0.10 | $5.00 | $2.00 | 200K | $3.50 |
USD per 1M tokens from Anthropic’s official pricing page as of August 31, 2026. Blended assumes a 3:1 input:output ratio. Get keys at the Anthropic console.
Claude pricing is the most legible of the big three: Haiku 4.5 at $1 input / $5 output per 1M tokens, Sonnet 5 at $2 / $10, Opus 5 at $5 / $25, and the Mythos-class Fable 5 at $10 / $50. Each step buys roughly a class of capability; all current models offer at least 200K context, with the frontier models at 1M.
Claude is also the only one of the big three with no long-context surcharge: the full 1M-token window on current models bills at standard rates, where GPT charges 2× input beyond 272K tokens and Gemini 2× beyond 200K. For huge-prompt workloads, that quietly flips the price comparison.
Prompt caching is explicit rather than automatic: you mark cache breakpoints, pay 1.25× input to write a prefix into the five-minute cache (2× for the one-hour cache), then read it back at a tenth of the fresh-input price. For long system prompts and tool definitions, reads dominate writes and the net saving is dramatic.
The Batch API halves both input and output prices for asynchronous jobs, and it stacks with cache pricing — batched, cached input on Sonnet 5 costs pennies per million tokens.
Claude bills per token with separate input and output rates per million tokens: Haiku 4.5 at $1 / $5, Sonnet 5 at $2 / $10, Opus 5 at $5 / $25, and Claude Fable 5 at $10 / $50. Cache reads cost about a tenth of fresh input on every model.
Claude Haiku 4.5 at $1 input / $5 output per 1M tokens ($2/1M blended). It supports the same tool use, vision, and reasoning surface as its bigger siblings with a 200K context window — the default pick for high-volume production traffic on Claude.
You set cache breakpoints on stable prompt prefixes. Writing a prefix to the cache costs a small premium over the normal input rate; every subsequent request that reuses it pays the cache-read rate — roughly 10× cheaper — until the cache expires. Apps with long system prompts, RAG contexts, or big tool definitions save the most.
Slightly, tier for tier: Opus 5 ($10/1M blended) vs GPT-5.6 Sol ($8), Sonnet 5 ($4) vs Terra ($4.50) — effectively even in the middle — and Haiku 4.5 ($2) above Luna ($0.45) at the small end. But on prompts past ~272K tokens the comparison flips, because GPT bills 2× input beyond that threshold and Claude doesn’t. The matchup table below puts all three vendors side by side.
No — API usage is pay-as-you-go. Claude.ai consumer subscriptions (Pro, Max) are a separate product and include no API credits.
Also see: OpenAI API pricing · Gemini API pricing · Grok API pricing · DeepSeek API pricing · Kimi API pricing · GLM API pricing · MiniMax API pricing · MiMo API pricing · all 58 models
ReqKey gives every key a credit balance — validate, deduct, enforce limits, and watch per-consumer Claude spend in real time.