Skip to content
Claude News

Guide

Claude models

Which Claude models are current and how they differ

Updated 21 October 202636 stories

In short

The current Claude lineup is the 5.5 family plus Fable 5.1 on top. Claude Opus 5.5 (claude-opus-5-5) costs $4 and $20 per million input and output tokens, and Anthropic says it delivers Fable 5.1-level performance on most work. Sonnet 5.5 is the default Sonnet on the Anthropic API, with 1M context at $2/$10 per Mtok. Haiku 5.5 charges $0.10 input and $0.50 output per million tokens on prompts under 100k. Claude Fable 5.1 runs 1M context at $10/$50 per Mtok, with $0.25/Mtok cache reads. Mythos 5.1 is the same model with more permissive safeguards, and only vetted cyberdefenders and life scientists can get it.

The most recent confirmed change is in pricing. Along with the Haiku 5.5 launch, Anthropic halved cache reads on Sonnet 5.5, from $0.20 to $0.10 per million tokens, and says that makes most agentic work around 20% cheaper. The same post says the updated tokenizer uses slightly more tokens per task. Token volume also matters at max effort: Haiku 5.5 generated 440M output tokens on the Artificial Analysis index, and thinking tokens are billed as output.

Not confirmed: Claude Fable 5.5. A LiteLLM entry listed a 1,000,000-token input window, and a leaked Azure screenshot showed $10 and $50 per million tokens, which TestingCatalog thinks may just be carried over. Anthropic hasn’t confirmed any of it. Security-heavy work still needs planning. On Opus 5.5, most cybersecurity tasks get rerouted to Opus 4.8.

At a glance

ModelIDStatusAs ofSource
Claude Opus 5.5claude-opus-5-5Released September 22, 2026. $4/$20 per million input/output tokens, cache reads $0.20 per million. Live on all platforms, including AWS, Google Cloud and Azure. Most cybersecurity tasks rerouted to Opus 4.8.22 September 2026Read
Claude Sonnet 5.5Released. Cache reads halved from $0.20 to $0.10 per million tokens.7 October 2026Read
Claude Haiku 5.5Released. On prompts under 100k: $0.10 per million input tokens, $0.50 per million output. Prompts over 100k get a smaller cut. First Haiku with an effort setting.7 October 2026Read
Claude Fable 5.1claude-fable-5-1Released on the Claude API, AWS, Google Cloud and Azure. Cache reads $0.25 per million tokens, rest of pricing unchanged from Fable 5.1 September 2026Read
Claude Mythos 5.1Same model as Fable 5.1 with more permissive safeguards. Vetted cyberdefenders and life scientists only, through two trusted access programs (currently a set of US organizations).1 September 2026Read
Claude Fable 5.5Unreleased and unconfirmed by Anthropic. LiteLLM entry removed. A leaked Azure screenshot showed $10 and $50 per million input/output tokens.17 October 2026Read
Claude Opus 5claude-opus-5Released July 2026. $5/$25 per million input/output tokens. Fast mode about 2.5x faster at double the price. Default on Claude Max at launch.24 July 2026Read
Claude Sonnet 5Generally available to all users via the Claude interface and API.1 July 2026Read
Claude Fable 5Released June 2026. $10/million input tokens, $50/million output. 1M context, 128k output limit. Adaptive thinking on by default.9 June 2026Read
Claude Mythos 5Released June 2026. Matches Fable 5's capabilities. Exclusive to Project Glasswing participants. $10/million input tokens, $50/million output. 1M context, 128k output limit.9 June 2026Read
Claude Opus 4.8Released May 28, 2026. Pricing unchanged. Fast mode runs 2.5x faster and costs 3x less, via /fast in Claude Code, with API users on a waitlist.28 May 2026Read

Milestones

  1. Anthropic confirms plans to release its Mythos vulnerability-discovery models to the public, with no date given.
  2. Claude Opus 4.8 launches with unchanged pricing and a Fast mode.
  3. Fable 5 and Mythos 5 launch at $10/$50 per million tokens, and Mythos 5 is limited to Project Glasswing participants.
  4. Anthropic temporarily disables Fable 5 and Mythos 5 for all users after a US Commerce Department licensing letter.
  5. Sonnet 5 becomes generally available, and Anthropic publishes its System Card.
  6. Claude Opus 5 ships at $5/$25 and becomes the default on Claude Max.
  7. Biology refusals on Claude Fable 5 fall roughly 85% after a safeguard retune.
  8. A statistical text watermark goes live on Claude models released after August 2, 2026.
  9. Fable 5.1 ships with cache reads cut 75% to $0.25 per million tokens, and Mythos 5.1 goes to vetted defenders and scientists.
  10. Claude Code 2.1.257 makes Fable 5.1 the default Fable model.
  11. Claude Opus 5.5 launches at $4/$20 with cache reads at $0.20 per million tokens.
  12. Claude Code 2.1.280 makes Opus 5.5 the default on Pro and Team Standard plans.
  13. Sonnet 5.5 ships at $2/$10 per Mtok and becomes the default Sonnet on the Anthropic API.
  14. Haiku 5.5 launches at $0.10/$0.50 on short prompts, and cache reads on Sonnet 5.5 are halved.
  15. An unconfirmed Claude Fable 5.5 model ID appears in LiteLLM’s Azure config.

FAQ

What is the latest Claude model?

The confirmed releases are the 5.5 family. Opus 5.5 came out in September 2026, Sonnet 5.5 followed later that month, and Haiku 5.5 arrived in October 2026. Fable 5.1 sits above them at $10/$50 per Mtok.

How much does Claude Opus 5.5 cost?

Opus 5.5 costs $4 per million input tokens and $20 per million output, with cache reads at $0.20 per million. On Opus 5 those were $5, $25 and $0.50. Anthropic puts the saving at 40% on typical workloads.

Has Claude Fable 5.5 been released?

No, not as of October 2026. A LiteLLM commit added azure_ai/claude-fable-5-5, but the entry was removed after about a day. A leaker says partners already have the model and expects a release next week. Anthropic hasn’t confirmed anything.

What is the difference between Claude Fable 5.1 and Mythos 5.1?

Mythos 5.1 is the same model as Fable 5.1 with more permissive safeguards. Only vetted cyberdefenders and life scientists can use it, through two trusted access programs that currently cover a set of US organizations.

Why do Claude 5.5 models hand security tasks to older models?

The safeguards do the rerouting. On Opus 5.5, most cybersecurity tasks get rerouted to Opus 4.8. Sonnet 5.5 is the first Sonnet with cyber safeguards, so higher-risk security tasks visibly fall back to Sonnet 5.

Latest