Cutoff Data
AI model price tracker
What the major models actually cost per million tokens, derived from the billed cost of real API calls we make every hour — not from vendor rate cards.
Last reading 10/8/2026, 7:17:04 AM UTC
| Model | Provider | Input / 1M | Output / 1M | Context | Status |
|---|---|---|---|---|---|
| Gemini 2.5 Flash Litegoogle/gemini-2.5-flash-lite | $0.10 | $0.40 | 1M | Not responding | |
| Gemini 3.1 Flash Litegoogle/gemini-3.1-flash-lite | $0.25 | $1.50 | 1M | Responding | |
| Gemini 2.5 Flashgoogle/gemini-2.5-flash | $0.30 | $2.50 | 1M | Not responding | |
| Gemini 2.5 Progoogle/gemini-2.5-pro | $1.25 | $10.00 | 1M | Not responding | |
| Gemini 3.5 Flashgoogle/gemini-3.5-flash | $1.50 | $9.00 | 1M | Not responding | |
| Gemini 3.1 Pro (preview)google/gemini-3.1-pro-preview | $2.00 | $12.00 | 1M | Responding | |
| Gemini 3.6 Flashgoogle/gemini-3.6-flash | — | — | 1M | Not responding | |
| Gemini 3.8 Flashgoogle/gemini-3.8-flash | — | — | 1M | Responding | |
| GPT-5 Nanoopenai/gpt-5-nano | OpenAI | — | — | — | Not responding |
| GPT-5 Miniopenai/gpt-5-mini | OpenAI | — | — | — | Not responding |
| GPT-5.6 Lunaopenai/gpt-5.6-luna | OpenAI | — | — | — | Responding |
| GPT-5.6 Terraopenai/gpt-5.6-terra | OpenAI | — | — | — | Responding |
| GPT-5.4 Miniopenai/gpt-5.4-mini | OpenAI | — | — | — | Responding |
| GPT-5.6 Solopenai/gpt-5.6-sol | OpenAI | — | — | — | Responding |
Recent changes
- 2026-11-07GPT-5.6 SolGPT-5.6 Sol responded again after a failed check
- 2026-11-07GPT-5.6 SolGPT-5.6 Sol stopped responding (timeout)
- 2026-11-03GPT-5.6 SolGPT-5.6 Sol responded again after a failed check
- 2026-11-03GPT-5.6 SolGPT-5.6 Sol stopped responding (timeout)
- 2026-10-31GPT-5.4 MiniGPT-5.4 Mini responded again after a failed check
- 2026-10-31GPT-5.6 SolGPT-5.6 Sol responded again after a failed check
- 2026-10-31GPT-5.6 TerraGPT-5.6 Terra responded again after a failed check
- 2026-10-31GPT-5.6 LunaGPT-5.6 Luna responded again after a failed check
- 2026-10-31Gemini 3.1 Pro (preview)Gemini 3.1 Pro (preview) responded again after a failed check
- 2026-10-31Gemini 3.1 Flash LiteGemini 3.1 Flash Lite responded again after a failed check
- 2026-10-31Gemini 3.8 FlashGemini 3.8 Flash responded again after a failed check
- 2026-10-28GPT-5.4 MiniGPT-5.4 Mini stopped responding (credit_limit)
- 2026-10-28GPT-5.6 SolGPT-5.6 Sol stopped responding (credit_limit)
- 2026-10-28GPT-5.6 TerraGPT-5.6 Terra stopped responding (credit_limit)
- 2026-10-28GPT-5.6 LunaGPT-5.6 Luna stopped responding (credit_limit)
- 2026-10-28Gemini 3.1 Pro (preview)Gemini 3.1 Pro (preview) stopped responding (credit_limit)
- 2026-10-28Gemini 3.1 Flash LiteGemini 3.1 Flash Lite stopped responding (credit_limit)
- 2026-10-28Gemini 3.8 FlashGemini 3.8 Flash stopped responding (credit_limit)
- 2026-10-27GPT-5.4 MiniGPT-5.4 Mini responded again after a failed check
- 2026-10-27GPT-5.6 SolGPT-5.6 Sol responded again after a failed check
- 2026-10-27GPT-5.6 TerraGPT-5.6 Terra responded again after a failed check
- 2026-10-27GPT-5.6 LunaGPT-5.6 Luna responded again after a failed check
- 2026-10-27Gemini 3.1 Pro (preview)Gemini 3.1 Pro (preview) responded again after a failed check
- 2026-10-27Gemini 3.1 Flash LiteGemini 3.1 Flash Lite responded again after a failed check
- 2026-10-27Gemini 3.8 FlashGemini 3.8 Flash responded again after a failed check
How this is measured
Every hour we send an identical short prompt to each tracked model through a single gateway account and record what the call was billed. Dividing the billed input and output cost by the tokens consumed gives an observed price per million tokens.
Prices shown are therefore what one buyer was actually charged, at that hour, on that route. Enterprise agreements, batch discounts and cached-input rates will differ. Where a provider does not itemise cost, the price cell reads “—”.
A “not responding” model returned an error on the most recent sweep. That is a fact about our route to it, not necessarily a global outage.