Anthropic Claude model family (as of 2026-09-30)

Anthropic builds the Claude family of large language models. Since mid-2026 the lineup has four generally available tiers (Fable, Opus, Sonnet, Haiku) plus an invite-only tier (Mythos).

Current lineup (generally available)

ModelAPI IDReleasedContext / max outputNote
Claude Fable 5.1claude-fable-5-12026-09-011M / 128Kdemanding reasoning, long-horizon agentic work
claude-opus-5claude-opus-5-52026-09-221M / 128Klong-running agentic coding and knowledge work; Anthropic’s suggested default
claude-sonnet-5claude-sonnet-5-52026-09-281M / 128Kbest speed/intelligence mix
Claude Haiku 4.5claude-haiku-4-5-202510012025-10-15200K / 64Kfastest, near-frontier; oldest model still listed as current

Invite-only: Claude Mythos 5.1 (claude-mythos-5-1, 2026-09-01), same model as Fable 5.1, Project Glasswing participants only.

Tiers, as Anthropic positions them

  • Fable: the top tier. A “Mythos-class” model released publicly with safety classifiers (cyber, bio/chem, distillation) that can decline requests. Slower, most expensive.
  • Mythos: the same underlying model with those safeguards relaxed, for vetted defensive-cybersecurity (Project Glasswing) and life-science users. Not generally available.
  • Opus: long-running agentic coding and knowledge work; middle of the price range. Anthropic says start here for most workloads and move to Fable when Opus at higher effort falls short.
  • Sonnet: speed and intelligence balance, cheaper and faster than Opus.
  • Haiku: fastest and cheapest.

How to choose

Start with Opus 5.5; use Sonnet 5.5 when latency/cost matter; Haiku 4.5 for high-volume, low-latency work (200K context, manual extended thinking, no effort parameter); Fable 5.1 when Opus 5.5 at higher effort still fails your evals. All current models take text and images and output text; current 5.x models use adaptive thinking, and non-default temperature/top_p/top_k are rejected from Opus 4.7 onward.

Legacy (still served) and retired

StatusModels
Legacy, activeFable 5 (not before 2027-06-09), claude-opus-5 (2027-07-24), claude-opus-4-dot-8 (2027-05-28), claude-opus-4-dot-7 (2027-04-16), Opus 4.6 (2027-02-05), Opus 4.5 (2026-11-24), claude-sonnet-5 (2027-06-30), Sonnet 4.6 (2027-02-17)
DeprecatedSonnet 4.5 (deprecated 2026-09-30, retires 2026-11-30); claude-mythos-preview (deprecated 2026-06-09, retirement TBA)
RetiredOpus 4.1 (2026-08-05), Opus 4 and Sonnet 4 (2026-06-15), Haiku 3 (2026-04-20), Haiku 3.5 and Sonnet 3.7 (2026-02-19), Opus 3 (2026-01-05), Sonnet 3.5 (2025-10-28)

Dates in parentheses are “not sooner than” retirement commitments on Anthropic-operated platforms (Claude API, Claude Platform on AWS, Microsoft Foundry); Bedrock and Google Cloud set their own. Anthropic gives at least 60 days’ notice before retiring a public model.

Deprecation calendar (next months)

DateEvent
2026-10-15Earliest possible retirement of Haiku 4.5 (no deprecation notice issued yet; with the 60-day notice policy, actual retirement cannot occur before roughly late Nov 2026 unless notice is given: inference)
2026-11-24Earliest possible retirement of Opus 4.5 (no notice yet)
2026-11-30Sonnet 4.5 retires (replacement: claude-sonnet-5-5)
2027-02 onwardOpus 4.6 (02-05), Sonnet 4.6 (02-17) earliest retirement

Claude products

Desktop app: Claude Desktop.

Claude models are served through the Claude API / Claude Platform, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry and Claude Platform on AWS; consumer/enterprise access via claude.ai and the Claude apps; and Claude Code for coding (Fable 5.1 announcement lists claude.ai, Claude Code and Enterprise).

Other Claude 4-era notes

Sources