Claude Models Explained

The Current Lineup

Four models make up the generally available comparison set in September 2026: Claude Fable 5.1 at the top, Claude Opus 5 as the recommended default, Claude Sonnet 5 as the balance point, and Claude Haiku 4.5 as the fast, cheap tier. Claude Mythos 5.1 exists alongside Fable 5.1 but is not a public option.

Fable 5.1 shipped on September 1, 2026 and replaced Fable 5 as the most capable generally available model. Opus 5 shipped on July 24, 2026 and replaced Opus 4.8. Any guide still naming Fable 5 as the flagship or Opus 4.8 as the current Opus is out of date.

API IDs, Prices, And Limits

Model IDs and token prices come from the Models overview and the Platform pricing page. Prices are per million tokens (MTok), input first and output second. Context and maximum output are the documented per-request ceilings, not guarantees for any given plan.

Cache reads are normally 10% of the base input price, but Fable 5.1 and Mythos 5.1 price them at 2.5% — $0.25 per million tokens. The Batch API remains 50% off for asynchronous work on supported models.

• Claude Fable 5.1 — `claude-fable-5-1` — $10 / $50 — 1M context — 128K max output — adaptive thinking always on — default effort high — knowledge cutoff June 2026 — retirement not sooner than September 1, 2027.
• Claude Opus 5 — `claude-opus-5` — $5 / $25 — 1M context — 128K max output — adaptive thinking — default effort high — knowledge cutoff May 2026 — retirement not sooner than July 24, 2027.
• Claude Sonnet 5 — `claude-sonnet-5` — $2 / $10 — 1M context — 128K max output — adaptive thinking — default effort high — knowledge cutoff January 2026 — retirement not sooner than June 30, 2027.
• Claude Haiku 4.5 — `claude-haiku-4-5`, dated alias `claude-haiku-4-5-20251001` — $1 / $5 — 200K context — 64K max output — extended thinking — knowledge cutoff February 2025 — retirement not sooner than October 15, 2026.
• Claude Mythos 5.1 — same underlying model as Fable 5.1 with more permissive safeguards — $10 / $50 — trusted access through Project Glasswing only.

How To Choose

Anthropic’s own advice is unusually direct: start with Opus 5. Reach for Fable 5.1 when the work is demanding reasoning or long-horizon agentic execution, or when your evals on Opus 5 at higher effort still fall short. Drop to Sonnet 5 when quality is adequate and you want the speed and price. Use Haiku 4.5 when the task is narrow and the volume is large.

Choose by failure cost first. If a wrong or incomplete answer is expensive, start high and measure down. For developers, test model choice against real evals rather than vibes, using the same prompt, tools, context size, retry rules, and stop handling production will use. A model that is cheaper per token can still be more expensive if it retries, over-searches, or needs human repair.

Fable 5.1’s cache-read discount changes that arithmetic for repetitive agentic workloads. Anthropic reports roughly 25% savings on typical workloads and up to about 45% on highly agentic ones compared with Fable 5, so a Fable-versus-Opus cost comparison should model cache hits rather than list price alone.

Legacy Models And Mythos

Older models remain listed and callable: Fable 5, Opus 4.8, 4.7, 4.6, and 4.5, and Sonnet 4.6 and 4.5. They are useful for pinned production systems and reproducibility, but they are not the right starting point for new work. Check the retirement dates on the Models overview before pinning a dated ID.

Mythos 5.1 is not a normal public tier. Anthropic describes it as the same underlying model as Fable 5.1 with more permissive safeguards, available only through trusted access and Project Glasswing, including a life-sciences path that is enrolling its first participants. Treat it as a restricted capability, not a model you can select.

What Anthropic Published About Fable 5.1

The Fable 5.1 announcement includes a benchmark table comparing Fable 5.1 against Fable 5 and Opus 5. These are Anthropic’s own reported figures, not independent measurements, and should be cited as such. For third-party benchmark analysis, see claudereports.com rather than treating vendor numbers as settled.

Selected reported results: Terminal-Bench-Science 0.1 at 52.6% for Fable 5.1 versus 29.0% for Opus 5; Terminal-Bench 4.0 at 55.8% (Mythos 5.1 at 60.9%) versus 52.3%; GDPval-AA v2 at 1853 versus 1824; OSWorld 2.0 at 77.9% partial and 41.7% strict; Humanity’s Last Exam at 60.9% without tools and 65.0% with tools; AutomationBench at 31.4%; CursorBench 3.2.0 at 73.4%.

Pricing Implications

For API work, output tokens usually dominate cost, so concise output schemas and structured tool results matter more than shaving input. Prompt caching is the largest single lever for repeated context, and it is disproportionately valuable on Fable 5.1 because of the 2.5% cache-read rate.

For consumer plans, model access is constrained by plan, capacity, and product policy rather than raw API pricing. Opus 5 is the default model on Claude Max and the strongest model available on Claude Pro. Verify current plan-level model access on Anthropic’s pricing and Help Center pages before committing.