price of compute

The price of a token

API list prices for frontier models, in USD per million tokens — standard tier, text, base context. Each number is verified against the lab’s official pricing page (date shown); the daily series is in the free API. More labs get added as they’re verified — never before.

All labsAnthropicDeepSeekGoogleMistralOpenAIxAI
Input vs output $/MTok · log-log · 5×/6× guides
$0.3$1$0.1$0.3input →output ↑5×6×deepseek-v4-prodeepseek-v4-flash

The dashed rays are constant output÷input ratios — nearly the whole market prices output at 5–6× input. Distance along a ray is price class; distance off it is pricing strategy.

The ladder · every model, one scale
$0.1$0.3$1deepseek-v4-flash: $0.14 in · $0.28 out per MTokdeepseek-v4-flashdeepseek-v4-pro: $0.43 in · $0.87 out per MTokdeepseek-v4-proinputoutput · $/MTok, log scale

DeepSeek

ModelInput $/MTokOutput $/MTokOut÷InBlended 3:1Verified
deepseek-v4-pro$0.43$0.872.0×$0.542026-08-11
deepseek-v4-flash$0.14$0.282.0×$0.182026-08-11

Last snapshot: 2026-08-11 06:05 UTC · List prices, not negotiated rates; cached-input, batch, and long-context tiers vary by lab and are noted where they differ. Same rule as the GPU side: a missing number beats a wrong one.