LLM API pricing tracker
What every major model costs per million tokens, checked against the vendors' own pricing pages. Standard tier, USD, cache and long-context rules noted.
Highest output price per 1M (Claude Fable 5.1)
Cheapest frontier-class output (Claude Sonnet 5.5)
Models tracked · 13 read from vendor pages
| Model | Input / 1M | Output / 1M | Cached input | Context | Verified |
|---|---|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | $0.25 | 1M | 2026-11-08 ● primary source |
| Claude Opus 5.5 | $4.00 | $20 | $0.20 | 1M | 2026-11-08 ● primary source |
| Claude Sonnet 5.5 | $2.00 | $10 | $0.10 | 1M | 2026-11-08 ● primary source |
| Claude Haiku 5.5 (prompt ≤100K) | $0.10 | $0.50 | $0.01 | tiered | 2026-11-08 ● primary source |
| Claude Opus 5 / 4.8 / 4.7 / 4.6 | $5.00 | $25 | $0.50 | 1M | 2026-11-08 ● primary source |
| Claude Sonnet 4.6 | $3.00 | $15 | $0.30 | 1M | 2026-11-08 ● primary source |
| GPT-6 Astra | $10 | $50 | $1.00 | 1.05M | 2026-11-08 ● primary source |
| GPT-5.6 Sol | $4.00 | $20 | — | — | 2026-11-08 ● secondary, pending re-check |
| GPT-5.6 Terra | $2.00 | $12 | — | — | 2026-11-08 ● secondary, pending re-check |
| GPT-5.6 Luna | $0.20 | $1.20 | — | — | 2026-11-08 ● secondary, pending re-check |
| GPT-5.4 mini | $0.75 | $4.50 | $0.07 | — | 2026-11-08 ● secondary, pending re-check |
| GPT-5.4 nano | $0.20 | $1.25 | $0.02 | — | 2026-11-08 ● secondary, pending re-check |
| Gemini 4 Argon | $2.00 | $10 | $0.10 | — | 2026-11-08 ● primary source |
| Gemini 3.8 Flash | $0.75 | $3.75 | $0.07 | — | 2026-11-08 ● primary source |
| Gemini 3.6 Flash | $0.75 | $3.75 | $0.07 | — | 2026-11-08 ● primary source |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | $0.03 | — | 2026-11-08 ● primary source |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.03 | — | 2026-11-08 ● primary source |
| Gemini 3.1 Pro Preview | $2.00 | $12 | $0.20 | tiered | 2026-11-08 ● primary source |
How to read this
Input vs output. You pay for the tokens you send (input) and the tokens the model writes back (output). Output is typically 5x the input rate, so verbose answers cost far more than long prompts.
Cached input. If you send the same long prefix repeatedly (a system prompt, a document), vendors charge a much lower rate for the repeated part, usually 2.5–10% of the input price. Anthropic charges a premium to write the cache; OpenAI charges 1.25x input for cache writes on GPT-6 Astra.
Batch. All three vendors offer roughly 50% off for asynchronous jobs that can wait minutes to hours.
Tiers and cliffs. Some models charge more above a prompt-length threshold (OpenAI above 272K input tokens, Gemini Pro above 200K, Claude Haiku 5.5 above 100K), and the surcharge applies to the whole request. Anthropic's current flagship tiers include the full 1M window at one price.
Introductory pricing. Google's Gemini 4 Argon and the 3.x Flash models carry published end dates for launch pricing. Build those dates into any cost forecast.
Tokenizers differ. Anthropic notes that Claude 4.7 and later produce roughly 30% more tokens for the same text than earlier models. Compare cost per task, not cost per token.
Common questions
What is the cheapest frontier-class model right now?
Among current flagship tiers, Anthropic's Claude Sonnet 5.5 ($2 / $10) and Google's Gemini 4 Argon ($2 / $10 until November 30, 2026) are the lowest list prices. Argon doubles on December 1.
Why do output tokens cost more than input tokens?
Generating text is sequential and compute-heavy, while reading a prompt can be processed in parallel. Vendors price output at roughly 5x input to reflect that.
How do I estimate my monthly API bill?
Multiply average input tokens per request by the input price, add average output tokens times the output price, divide by one million, then multiply by monthly requests. Apply cache and batch discounts where they fit your workload.
How often is this table updated?
Nightly. Rows marked as primary were read directly from the vendor's pricing page on the date shown; secondary rows come from sources that cite the vendor page and are re-checked until a primary read is possible.
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.