What changed in Claude: a dated log for developers
28 Sept 2026
New model
Same per-token prices and tokenizer as Sonnet 5, a lower cache minimum, and five breaking changes —
thinking: disabledand forced tool_choice both return a 400.22 Sept 2026
New model
Cheaper per token than Opus 5, with cache reads at a twentieth of input. Thinking cannot be turned off, and the effort default drops to medium.
22 Sept 2026
Plans
Five-hour usage limits raised on paid plans
Announced with Opus 5.5 for Pro, Max, Team and seat-based Enterprise, alongside a rate-limit reset subscribers can save and use when they choose.
16 Sept 2026
Plans
Claude Cowork became available from any conversation, with new artifact tools for designs, decks and documents across all plans.
1 Sept 2026
Pricing
Sonnet 5's launch price becomes its standard price
The price first announced as introductory through August 31 stays; the increase scheduled for September 1 does not take effect.
1 Sept 2026
New model
Claude Fable 5.1 and Mythos 5.1 released
Fable 5's successors at the same per-token price, with cache reads cut to a quarter of Fable 5's. Forced tool_choice returns a 400 and thinking blocks become bound to their conversation.
31 Aug 2026
API behaviour
Preserved thinking enforced for new accounts
Accounts created on or after this date get a 400 when a replayed thinking block follows an edit to earlier history, on the models that run the check.
5 Aug 2026
Retirement
Requests to it on the Claude API now fail; Amazon Bedrock and Google Cloud keep their own schedules.
24 Jul 2026
New model
Priced at half Fable 5's rate, with thinking on by default and a lower cache minimum than Opus 4.8.
19 Jul 2026
Plans
Fable 5's promotional inclusion on plans ends
A promotional period that included Fable 5 in plan usage ended. Under the current rules, Pro plans and standard Team seats reach Fable models through usage credits, while Max and premium seats can spend up to half of their weekly limits on them.
30 Jun 2026
New model
Launched at a price announced as introductory, with a new tokenizer that counts more tokens than Sonnet 4.6's.
15 Jun 2026
Retirement
Claude Opus 4 and Sonnet 4 retired
Both stopped answering on the Claude API; partner platforms set their own dates.
9 Jun 2026
New model
A new top tier above Opus, with thinking always on and a 30-day data-retention requirement.
What earns a line on this log
Anthropic publishes a great deal of news. This log keeps a narrower record: a change earns a line only if it moves what a developer pays, changes what a request is allowed to contain, or changes which model or plan a team can use. A new model qualifies; so does a price that didn't change when it was scheduled to, a model retirement, a shift in what a subscription includes, or a new enforcement rule that turns a harmless request into an error. Benchmark results and launch adjectives don't, because neither is something you can act on from a reference page.
How each line is recorded
Every entry carries the date Anthropic gave, the primary source it came from, and a link to the page on this site that works through the consequence — the migration diff, the error fix, the plan explainer. Where a source gives an effective date rather than an announcement date, the effective date is used, and the note under the log says so. Entries render from the same facts module as every price and limit on the site, so the log cannot contradict the tables it points to: a model launch recorded here and the model's own reference page read from one record.
Why a log, when every page is already current
Every other page on this site shows the present state — the current rate, the current limits. That is the right default for a lookup, and the wrong one for someone returning after a few weeks who needs to know what moved since they last checked. A dated log answers that directly, and it makes a change that happened quietly — a price that stayed put, an enforcement date for new accounts — as visible as a launch.
How it stays honest
The log is part of the facts module, so it is held to the same rule as every price on the site: the build refuses to deploy once any entry is past its freshness limit, which forces a re-read of the sources rather than letting the page age quietly. See how we keep this accurate for that discipline, and the all-models pricing table for where the rates stand today.
Verified 2026-09-30 against Anthropic's release notes, newsroom, pricing and model-deprecation pages (canonical: https://support.claude.com/en/articles/12138966-release-notes, https://www.anthropic.com/news, https://platform.claude.com/docs/en/about-claude/pricing, https://platform.claude.com/docs/en/about-claude/model-deprecations).Could not confirm: Where Anthropic gives an effective date rather than an announcement date — Sonnet 5's price decision is the example — the effective date is shown.