Claude Models
Anthropic's lineup as of November 2026: Fable 5.1 for the hardest work, Opus 5.5 as the default, Sonnet 5.5 for everyday work at half Opus's price, and Haiku 5.5 for high volume.
Current lineup
| Model | ID | Released | Price per 1M tokens, in / out | Use it for |
|---|---|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 |
2026-09-01 | $10 / $50 | The hardest reasoning and long-running agent work |
| Claude Opus 5.5 | claude-opus-5-5 |
2026-09-22 | $4 / $20 | The default for coding and agents |
| Claude Sonnet 5.5 | claude-sonnet-5-5 |
2026-09-28 | $2 / $10 | Everyday coding, agent and enterprise work at speed |
| Claude Haiku 5.5 | claude-haiku-5-5 |
2026-10-07 | $0.10 / $0.50 | High-volume work and sub-agents |
All four take 1M tokens of context and return up to 128K output tokens. Haiku 5.5's price covers prompts up to 100K tokens and rises to $0.50 / $2.50 beyond that. Fable 5.1 requires 30-day data retention. Source: claude-api skill, model table cached 2026-10-06; Source: Anthropic model docs, checked 2026-11-03
Claude Mythos 5.1 (claude-mythos-5-1) has Fable 5.1's capabilities and price, and is available only to organizations verified through Anthropic's verification programs, such as the Cyber Verification Program. Claude Mythos Preview, and Mythos 5 at launch, were limited to Project Glasswing partners. Source: Project Glasswing Source: Anthropic Mythos 5.1 overview, checked 2026-11-03
Earlier models are still served: Fable 5, Opus 5 ($5 / $25), Sonnet 5, Opus 4.8, 4.7, 4.6 and 4.5, Sonnet 4.6 and Haiku 4.5. Sonnet 4.5 is deprecated and retires on 2026-11-30. Sonnet 4 and Opus 4 were retired on 2026-06-15, and the Claude 3.5 models are retired. Source: Anthropic model deprecations, checked 2026-11-03
How the line moved in 2026
- 2026-06-09: Fable 5 (generally available) and Mythos 5 (restricted), the first broad Mythos-class release (after the invitation-only Mythos Preview of 2026-04-07). Access was suspended on June 12 and restored on July 1. Claude Fable 5 covers both.
- 2026-07-24: Opus 5, close to Fable 5's intelligence at half its price. Source: X/@claudeai, 2026-07-24
- 2026-09-01: Fable 5.1 at Fable 5's price, with cache reads cut from $1 to $0.25 per 1M tokens.
- 2026-09-22: Opus 5.5, priced below Opus 5 at $4 / $20.
- 2026-09-28: Sonnet 5.5 at Sonnet 5's price.
- 2026-10-07: Haiku 5.5 at a tenth of Haiku 4.5's price.
Source: Anthropic model docs and announcements, checked 2026-11-03
What changed for prompts and code
Several habits written for older models now do nothing or return an error.
| Old habit | Now |
|---|---|
| "Think step by step", "don't overthink" | Thinking is always on for Fable 5.1 and Opus 5.5. Set effort from low to max; Opus 5.5 defaults to medium. |
A fixed thinking budget (budget_tokens) |
Returns a 400 on every 5.x model. Adaptive thinking plus effort replaced it. |
| Prefilling the assistant turn to force JSON | Returns a 400 from the 4.6 generation on. Use structured outputs (output_config.format). |
temperature, top_p, top_k |
Return a 400 on every model under API version 2026-10-21 (the anthropic-version header). On older API versions, Claude 4.7 and later models already reject non-default values. Source: Anthropic model deprecations, checked 2026-11-03 |
Forced tool use (tool_choice any or tool) |
Returns a 400 on Fable 5.1, Opus 5.5 and Sonnet 5.5. Use auto, name the tool in the prompt, and set strict: true. |
| Editing earlier turns in a long session | Invalidates the model's saved thinking. Harnesses append instead of rewriting history. |
| CRITICAL, MUST, step-by-step scripts | Current models follow plain instructions and do worse with over-specified steps. State the goal, the constraints and how to verify. |
Source: claude-api skill, model migration and prompt-audit guides, 2026-10-06 Claude Prompting Principles applies this to prompt writing, private page to JSON output and Tool-Calling Design to tools.
Choosing between them
Anthropic's guidance is to start with Opus 5.5 and, before building a cascade of cheaper models, measure the same model at lower effort: on the newest models, low effort often matches an older model at high effort, and one model keeps one prompt cache. Source: claude-api skill, cost guidance, 2026-10-06 Kevin's own rule still decides defaults: route by role, and promote a new model only after a measured run against the incumbent. Cross-Harness Model Routing
For a bounded choice inside code, such as a label, a route or a score, a decision model is faster and cheaper than any of these.
Timeline
- 2026-11-03 | Page created as the owner of the current Claude lineup: Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, with release dates, prices and the API habits they retired. Source: claude-api skill; Anthropic model docs