Guides

    Every Claude Model Release: Timeline and Changelog

    The latest Claude model, every release since 2025 with dates, model IDs, prices and retirement dates, from Anthropic's own docs. Last verified 16-11-2026.

    November 16, 202615 min read
    Share:

    Published 16-11-2026. Last verified on 16-11-2026 against Anthropic's models overview, model deprecations page and Claude Platform release notes. We re-check this timeline every month and after every Anthropic launch; if a month passes without a check, treat the dates below as possibly out of date.

    Quick answer: what is the latest Claude model?

    The latest Claude model is Claude Fable 5.2 (API ID claude-fable-5-2), launched on 09-11-2026 together with Claude Mythos 5.2, which is the same model offered only to verified organizations. For most work, Anthropic itself recommends starting with Claude Opus 5.5 (22-09-2026) and moving up to Fable or down to Sonnet 5.5 or Haiku 5.5 as the job requires. This page is for developers and AI leads who run Claude in production and need one place to see what is current, what changed per release and which model IDs are about to stop working. The most urgent date right now: claude-sonnet-4-5-20250929 retires on 30-11-2026, so any code still calling it needs to move this month.

    How we keep this timeline honest

    Our method in three lines: every date, model ID, context window and price comes from Anthropic's own documentation, read on 16-11-2026; benchmarks come from Anthropic's launch posts and are labelled vendor-reported; and we re-check the page monthly. Our method page on how we evaluate skills and models explains the wider approach.

    Our gate for this page: a release only gets an entry if an Anthropic primary source names the model and gives a date. That sounds strict, and it is meant to be. The gate removed three things that other timelines in the search results include. First, rumoured release dates for the next model: none of the ones we saw on 16-11-2026 linked to an Anthropic source, so they are out. Second, dates from third-party timeline sites that list releases without citing Anthropic; we used none of their dates. Third, release dates guessed from retirement commitments: the deprecations page lists "not sooner than" dates exactly one year after most launches, which is tempting to read backwards, but we only use the launch dates from the release notes.

    We did not test the models ourselves for this page. Every benchmark below is Anthropic's own number; we would not compare it one-to-one with another vendor's table.

    The current lineup on 16-11-2026

    Anthropic's models overview lists four "latest" models. All four have a 1M-token context window, 128K tokens of maximum output and a reliable knowledge cutoff of June 2026. Prices are list prices per million tokens (MTok), input and output.

    ModelAPI IDReleasedPrice (input and output per MTok)Anthropic's positioning
    Claude Fable 5.2claude-fable-5-209-11-2026$8 and $40Demanding reasoning, long-horizon agentic work
    Claude Opus 5.5claude-opus-5-522-09-2026$4 and $20Long-running agentic coding and knowledge work
    Claude Sonnet 5.5claude-sonnet-5-528-09-2026$2 and $10Speed and intelligence combined
    Claude Haiku 5.5claude-haiku-5-507-10-2026From $0.10 and $0.50High-volume classification, extraction, routing

    The order in this table is the order of capability as Anthropic describes it, and here is why each sits where it does. Fable 5.2 is first because Anthropic's launch post shows it ahead of Opus 5.5 on almost every benchmark it reports, at twice the price. Opus 5.5 is second, and it is the one we would make the default for most teams, because Anthropic calls it the "strongest daily driver" and recommends starting there. Sonnet 5.5 is third: half the Opus price and faster. In our reading it is the right model for well-scoped everyday tasks, and it is where we would send most routine skill runs. Haiku 5.5 is fourth on capability but first on cost per job, which is what matters for bulk work.

    What would move them: a Fable price cut close to Opus would push Fable up to default for hard work; a Sonnet release that matches Opus on Anthropic's knowledge-work benchmarks would make Opus a niche choice. Our guide to the best Claude model for coding applies this to coding jobs specifically.

    If you want independent numbers next to Anthropic's, your agent can pull them in the same session where you decide. Through Agensi Tools, Artificial Analysis returns model details from $0.013 per call and model cost per task from $0.13 per call, and LMArena returns its text, code and vision leaderboards from $0.013 per call (starting prices read on the tools pages on 16-11-2026). You need no account or API key at either provider, your agent shows the price before it runs and a failed use is free. For skills that help you operate models in production, see the LLM ops collection.

    Every Claude model release, newest first

    This is the full list of model launches in the Claude Platform release notes from February 2025 to today. Context window and price are the values in the models overview on 16-11-2026, which for older models can differ from the launch-day values.

    DateModelAPI IDContextPrice per MTok (in and out)
    09-11-2026Fable 5.2 and Mythos 5.2claude-fable-5-2, claude-mythos-5-21M$8 and $40
    07-10-2026Haiku 5.5claude-haiku-5-51MFrom $0.10 and $0.50
    28-09-2026Sonnet 5.5claude-sonnet-5-51M$2 and $10
    22-09-2026Opus 5.5claude-opus-5-51M$4 and $20
    01-09-2026Fable 5.1 and Mythos 5.1claude-fable-5-1, claude-mythos-5-11M$10 and $50
    24-07-2026Opus 5claude-opus-51M$5 and $25
    30-06-2026Sonnet 5claude-sonnet-51M$2 and $10
    09-06-2026Fable 5 and Mythos 5claude-fable-5, claude-mythos-51M$10 and $50
    28-05-2026Opus 4.8claude-opus-4-81M$5 and $25
    16-04-2026Opus 4.7claude-opus-4-71M$5 and $25
    07-04-2026Mythos Preview (gated)claude-mythos-previewNot listedNot listed
    17-02-2026Sonnet 4.6claude-sonnet-4-61M$3 and $15
    05-02-2026Opus 4.6claude-opus-4-61M$5 and $25
    24-11-2025Opus 4.5claude-opus-4-5-20251101200K$5 and $25
    15-10-2025Haiku 4.5claude-haiku-4-5-20251001200K$1 and $5
    29-09-2025Sonnet 4.5claude-sonnet-4-5-20250929200K$3 and $15
    05-08-2025Opus 4.1claude-opus-4-1-20250805200K$15 and $75
    22-05-2025Opus 4 and Sonnet 4claude-opus-4-20250514, claude-sonnet-4-20250514200K$15 and $75, $3 and $15
    24-02-2025Sonnet 3.7claude-3-7-sonnet-20250219Not listedNot listed

    Three patterns stand out to us in this table, and each one changes how we would plan for 2027. The pace went up: 2025 had six launch dates in the release notes, 2026 has thirteen so far. Prices went down at the top: Opus dropped from $15 and $75 (Opus 4.1) to $4 and $20 (Opus 5.5), a cut of more than 70% on both input and output. And the context window became standard: every model launched since Opus 4.6 lists 1M tokens in the overview, against 200K for the 4.5 generation.

    The dated model IDs (with a date suffix such as 20251001) stopped with Opus 4.5. From Opus 4.6 onward, Anthropic uses IDs without a date, which matters if your code builds model IDs with a pattern. We would replace any such pattern with a plain list of allowed IDs that our own team reviews after each launch.

    What changed in the 2026 releases

    Release notes list every change; we keep only what changes a decision for your team. Headline numbers are vendor-reported, from Anthropic's launch posts.

    Fable and Mythos. Fable 5 (09-06-2026) introduced a tier above Opus, open to all customers, with Mythos as the same model for verified organizations. Fable 5.2 is, according to Anthropic, up to 30% faster than Fable 5.1 and costs $8 and $40 instead of $10 and $50. Its headline number in the Fable 5.2 launch post is 69.9% on FrontierSWE v2, against 62.3% for Opus 5.5 and 56.3% for Fable 5.1. Anthropic positions Fable 5.2 as an advisor or orchestrator that smaller models consult or receive work from; the agents and orchestration collection has skills for exactly that setup, and our Claude Fable 5 skills guide covers which skills hold up on it.

    Opus. Five releases this year: 4.6, 4.7, 4.8, 5 and 5.5. Opus 4.6 made adaptive thinking the recommended mode and deprecated manual thinking budgets. Opus 4.7 brought API breaking changes and a new tokenizer, per the release notes. Opus 5.5 is the big one for budgets: Anthropic says it performs at the level of Fable 5.1 on most work and costs 40% less to run than Opus 5 at default settings, with cache reads at $0.20 per million tokens.

    Sonnet. We see Sonnet as the release line with the biggest jump this year. Sonnet 5 launched at an introductory $2 and $10, made standard on 10-08-2026. Sonnet 5.5 kept that price and, in Anthropic's table, jumped from 10.3% to 70.6% on Terminal-Bench 4.0. On 07-10-2026 Anthropic halved Sonnet 5.5 cache reads to $0.10 per million tokens.

    Haiku. Haiku 5.5 is Anthropic's first Haiku with an adjustable effort setting and a 1M-token window, and Anthropic says it costs around 75% less to run than Haiku 4.5 on average. Its Terminal-Bench 4.0 score is 39.2% (vendor-reported), so it is a bulk and subagent model, not a coding lead. Watch the migration notes: code written for Haiku 4.5 can break, because sampling parameters, manual thinking budgets and assistant prefill can each return a 400 error, and the new tokenizer produces more tokens for the same text.

    The practical consequence, in our view: every model change is now also a code change, and we would never let a model swap ship without a code check. Model Migration Checker, $15, audits your code for hardcoded model versions, deprecated parameters and token limits before you swap. Model Migration Planner, also $15, builds a parity matrix, cost and latency projections and a staged rollout you can abort. The code quality and review collection has more checks of this kind.

    When was Claude Mythos released?

    Claude Mythos has four release dates, according to Anthropic's release notes. Mythos Preview was announced on 07-04-2026 as an invitation-only research preview for defensive cybersecurity work under Project Glasswing. Mythos 5 followed on 09-06-2026, Mythos 5.1 on 01-09-2026 and Mythos 5.2 on 09-11-2026, each alongside the Fable model of the same version.

    The difference with Fable is access, not capability: Anthropic states that Fable and Mythos are the same underlying model, and that Fable carries cybersecurity and biology safeguards so it can be offered to all users. Mythos 5.2 is for organizations verified through Anthropic's Cyber or Life Sciences Verification Programs. Mythos Preview is deprecated since 09-06-2026, with a retirement date still "to be announced" on the deprecations page. For nearly every team, Fable 5.2 is the version we would plan around; we would only apply for Mythos if your work is defensive security or biomedical research.

    Which Claude models are deprecated or retired?

    Anthropic's model deprecations page is the authority here, and it promises at least 60 days' notice before a public model retires. Requests to a retired model fail. Status on 16-11-2026:

    Model IDStatusRetirement dateAnthropic's replacement
    claude-sonnet-4-5-20250929Deprecated30-11-2026claude-sonnet-5-5
    claude-mythos-previewDeprecatedTo be announcedNot stated
    claude-opus-4-1-20250805Retired05-08-2026claude-opus-4-8
    claude-opus-4-20250514Retired15-06-2026claude-opus-4-8
    claude-sonnet-4-20250514Retired15-06-2026claude-sonnet-5-5
    claude-3-haiku-20240307Retired20-04-2026claude-haiku-4-5-20251001
    claude-3-5-haiku-20241022Retired19-02-2026claude-haiku-4-5-20251001
    claude-3-7-sonnet-20250219Retired19-02-2026claude-sonnet-5-5

    We have ordered this table by urgency, not by date. Sonnet 4.5 is first because it is the only public model with a fixed retirement date still ahead of us, two weeks from today. Mythos Preview is second because it is deprecated but without a date. The retired models follow, newest first, because those are the ones most likely to still sit in old scripts.

    Two other dates belong on your list. Fast mode for Opus 4.8 is removed on 30-11-2026, per the release notes of 30-10-2026. And API version 2026-10-21 removed the temperature, top_p and top_k parameters entirely; requests that send them on that version return a 400 error.

    One honest gap: the deprecations page lists claude-haiku-4-5-20251001 and claude-opus-4-5-20251101 as Active with "not sooner than" dates of 15-10-2026 and 24-11-2026. Neither has been deprecated so far, but both are the next likely candidates, so we would not start new work on either.

    To find who in your organization still depends on a retiring model, API Deprecation Deadline Ownership Watch, $1, tracks internal and external consumers exposed to upcoming API retirements. The APIs and backend collection covers the rest of the migration work.

    How do you check which Claude model you are on?

    Look at the model field in the API response: every Messages API response names the exact model that answered. For an organization-wide view, Anthropic's deprecations page describes the route we recommend first: open the Usage page in the Claude Console, click Export and read the CSV, which breaks usage down by API key and model. That shows within minutes which keys still call a deprecated ID.

    Three more checks, by where you work:

    1. In code: the Models API. First because it scales: GET /v1/models lists every model available to your key, with context window, maximum output and capabilities, so a CI job can fail when a pinned ID disappears.
    2. In Claude Code: the /model command. Second because it is where most developers switch models during a session; it shows and changes the active model. Our guide to updating Claude Code covers the release notes of the CLI itself, which is a separate changelog from the model releases on this page.
    3. In claude.ai: the model picker. Last because it only tells you what one person uses, not what your systems call.

    We would put the first check in your pipeline, because it runs on every build instead of when someone remembers. The DevOps and cloud collection has skills for wiring checks like this into CI.

    What is expected next?

    Reported, not confirmed: on 16-11-2026 we found no Anthropic announcement of a next model, so this block is empty by design. Rumours about future versions did not pass our gate.

    What we do know from Anthropic's own pages: the deprecations page commits Opus 5.5, Sonnet 5.5 and Haiku 5.5 to "not sooner than" September and October 2027, so they are safe to build on for at least a year. Anthropic's pace in 2026 has been more than one launch per month on average.

    We do not think anyone should check five pages by hand, and you do not have to. API Changelog Watcher, $7, flags when a provider quietly changes its docs. Your agent can also read the latest releases of Anthropic's public repositories, such as the SDKs and Claude Code, through the GitHub tool on Agensi Tools, which returns a repository and its releases from $0.0022 per call, without a GitHub account or API key. For broader monitoring skills, see the research and analysis collection.

    Your next step

    We sell skills and pay-per-call tools that your agent uses through one MCP connection, and we name ours in the sections above. No vendor paid to be listed.

    Our advice: do three things this week. Search your code and your Console usage export for claude-sonnet-4-5-20250929 and move it to claude-sonnet-5-5 before 30-11-2026. Decide which jobs stay on Opus 5.5, which move to Sonnet 5.5 or Haiku 5.5, and write it down; Route Each Task to the Right Model, $12, automates that split by sending easy work to a cheaper model and escalating only when a check fails. And before you switch a production workload to a new release, measure it on your own tasks: Agent Evals with Real Statistics, $12, sets up golden datasets and paired comparisons so you know whether the new model is better or just different. For current list prices per model, our Claude API pricing guide has the full table.

    To pull model data, leaderboards and repository releases into the same session where you plan, connect your agent once through Agensi MCP.

    Written by Samuel Rose, Founder of Agensi. Next scheduled review: 16-12-2026.