Open data

AI pricing changelog

Provider pricing pages have no history. When a price changes, the old number is overwritten and gone. This is the changelog those pages never shipped: every dated AI and LLM API pricing event we have on record, newest first, each one carrying the date it took effect and a link to the source it came from.

It is built from the open AI Price Index, covering 11 providers from March 2023 to July 2026. Two kinds of event appear. A repricing is a model whose published price moved, shown honestly as an increase or a decrease against the price it replaced. A new price is the day a model's published price first entered the record. We track the standard list price per million tokens, not promotional, batch, or cached rates. This page regenerates from the dataset on every build, so it stays current without anyone editing a list by hand.

Repricings
32
Increases / decreases
2 / 30
New prices logged
218
Span
March 2023 to July 2026
Kind
Provider

Showing 250 of 250 entries.

  • new price cache_read

    xAI grok-4.5 cache_read listed at $0.50 per million tokens

    Source (docs.x.ai)

  • new price input

    xAI grok-4.5 input listed at $2 per million tokens

    Source (docs.x.ai)

  • new price output

    xAI grok-4.5 output listed at $6 per million tokens

    Source (docs.x.ai)

  • decrease input

    Amazon nova-2-pro input $1.25 $1.25 per million tokens (down 0%)

    Source (aws.amazon.com)

  • decrease output

    Amazon nova-2-pro output $10 $10 per million tokens (down 0%)

    Source (aws.amazon.com)

  • decrease cache_read

    Anthropic mythos-5 cache_read $1 $1 per million tokens (down 0%)

    Source (docs.claude.com)

  • decrease cache_write_1h

    Anthropic mythos-5 cache_write_1h $20 $20 per million tokens (down 0%)

    Source (docs.claude.com)

  • decrease cache_write_5m

    Anthropic mythos-5 cache_write_5m $12.5 $12.5 per million tokens (down 0%)

    Source (docs.claude.com)

  • decrease input

    Anthropic mythos-5 input $10 $10 per million tokens (down 0%)

    Source (docs.claude.com)

  • decrease output

    Anthropic mythos-5 output $50 $50 per million tokens (down 0%)

    Source (docs.claude.com)

  • decrease cache_read

    xAI grok-4.20-0309-non-reasoning cache_read $0.20 $0.20 per million tokens (down 0%)

    Source (docs.x.ai)

  • decrease input

    xAI grok-4.20-0309-non-reasoning input $1.25 $1.25 per million tokens (down 0%)

    Source (docs.x.ai)

  • decrease output

    xAI grok-4.20-0309-non-reasoning output $2.5 $2.5 per million tokens (down 0%)

    Source (docs.x.ai)

  • increase input

    Mistral mistral-small-4 input $0.10 $0.15 per million tokens (up 50%)

    Source (mistral.ai)

  • increase output

    Mistral mistral-small-4 output $0.30 $0.60 per million tokens (up 100%)

    Source (mistral.ai)

  • new price input

    Mistral mistral-medium-3.5 input listed at $1.5 per million tokens

    Source (mistral.ai) - inferred

  • new price output

    Mistral mistral-medium-3.5 output listed at $7.5 per million tokens

    Source (mistral.ai) - inferred

  • new price cache_read

    xAI grok-4.20-multi-agent-0309 cache_read listed at $0.20 per million tokens

    Source (docs.x.ai)

  • new price input

    xAI grok-4.20-multi-agent-0309 input listed at $1.25 per million tokens

    Source (docs.x.ai)

  • new price output

    xAI grok-4.20-multi-agent-0309 output listed at $2.5 per million tokens

    Source (docs.x.ai)

  • new price input

    Mistral mistral-small-4 input listed at $0.10 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral mistral-small-4 output listed at $0.30 per million tokens

    Source (mistral.ai)

  • new price input

    xAI grok-4.20-0309-reasoning input listed at $1.25 per million tokens

    Source (docs.x.ai)

  • new price output

    xAI grok-4.20-0309-reasoning output listed at $2.5 per million tokens

    Source (docs.x.ai)

  • new price input

    AI21 jamba-mini input listed at $0.20 per million tokens

    Source (ai21.com) - inferred

  • new price output

    AI21 jamba-mini output listed at $0.40 per million tokens

    Source (ai21.com) - inferred

  • new price input

    Mistral devstral-2 input listed at $0.40 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral devstral-2 output listed at $2 per million tokens

    Source (mistral.ai)

  • new price input

    Mistral devstral-small-2 input listed at $0.10 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral devstral-small-2 output listed at $0.30 per million tokens

    Source (mistral.ai)

  • new price input

    Amazon nova-2-lite input listed at $0.30 per million tokens

    Source (aws.amazon.com)

  • new price output

    Amazon nova-2-lite output listed at $2.5 per million tokens

    Source (aws.amazon.com)

  • new price input

    Mistral ministral-3-14b input listed at $0.20 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral ministral-3-14b output listed at $0.20 per million tokens

    Source (mistral.ai)

  • new price input

    Mistral ministral-3-3b input listed at $0.10 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral ministral-3-3b output listed at $0.10 per million tokens

    Source (mistral.ai)

  • new price input

    Mistral ministral-3-8b input listed at $0.15 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral ministral-3-8b output listed at $0.15 per million tokens

    Source (mistral.ai)

  • new price input

    Mistral mistral-large-3 input listed at $0.50 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral mistral-large-3 output listed at $1.5 per million tokens

    Source (mistral.ai)

  • new price input

    Mistral codestral-25.08 input listed at $0.30 per million tokens

    Source (mistral.ai)

  • new price output

    Mistral codestral-25.08 output listed at $0.90 per million tokens

    Source (mistral.ai)

  • new price input

    AI21 jamba-large input listed at $2 per million tokens

    Source (ai21.com)

  • new price output

    AI21 jamba-large output listed at $8 per million tokens

    Source (ai21.com)

  • new price input

    meta-llama llama-4-maverick input listed at $0.27 per million tokens

    Source (together.ai) - inferred

  • new price output

    meta-llama llama-4-maverick output listed at $0.85 per million tokens

    Source (together.ai) - inferred

  • new price input

    meta-llama llama-4-scout input listed at $0.18 per million tokens

    Source (together.ai) - inferred

  • new price output

    meta-llama llama-4-scout output listed at $0.59 per million tokens

    Source (together.ai) - inferred

  • new price input

    Cohere command-r7b input listed at $0.037 per million tokens

    Source (cohere.com)

  • new price output

    Cohere command-r7b output listed at $0.15 per million tokens

    Source (cohere.com)

How to read this log

Each entry is a point-in-time fact: on the date shown, a provider's published price for that model and that token direction took the value shown, sourced to the page it came from. A repricing compares against the price it replaced in the record, so a price that went up reads as an increase, not a quiet down-and-to-the-right line. A new price is the first time we recorded a published number for that model, dated to its effective day. We log standard list prices in US dollars per million tokens. Archived entries are reconstructed from a dated snapshot (often the Wayback Machine) because the original page no longer carries the old number, which is the whole reason a changelog like this has to exist.

Why keep the receipts

If you want to know what a token actually cost the day you spent it, the live pricing page cannot tell you. The old number is gone the moment it changes. A dated, sourced log lets you reprice a historical bill at the rate that was really in effect, audit a "we cut prices" claim against the date the API page actually moved, and catch the increases that shipped with no announcement at all. Point-in-time repricing is also the engine behind Goei, our cost dashboard. For the wider argument about why the single "price of AI" stopped meaning anything, read the data story: AI prices did not fall, they fanned out.

Get the data

Open and CC BY 4.0. Found a wrong or missing price change? Send a pull request with the source. That is the point.

Citable: DOI 10.5281/zenodo.20730241. Prices validated as of 2026-07-10. Release v2026.07.10-cc35e0b.