O que é o Claude Opus 4.8?
Claude Fable 5 is Anthropic's first Mythos-class model released for general public access. Anthropic announced the model on June 9, 2026, alongside its sibling Claude Mythos 5. In Anthropic's own framing, "Fable 5's capabilities exceed those of any model we've ever made generally available," with the largest gains over Claude Opus 4.8 measured on long, multi-step autonomous tasks rather than on short single-turn benchmarks.
Fable 5 and Mythos 5 share the same underlying model. Mythos 5 is restricted to pre-approved organizations working on critical infrastructure; Fable 5 is the version available through the Claude API and through chat interfaces. In Anthropic's lineup, Fable 5 sits above Claude Opus 4.8 and Claude Sonnet 4.6 — all three are available on Overchat AI under the Pro subscription, without a separate Anthropic API key.
Claude Fable 5 capabilities — coding, vision, and long-context work
Fable 5's lead over Claude Opus 4.8 widens as task length increases. Anthropic states directly: "The longer and more complex the task, the larger Fable 5's lead over our other models." The model is designed to operate autonomously across hours of work without drifting from the original objective. Stripe used Fable 5 to migrate a 50-million-line Ruby codebase in approximately one working day — a project estimated at two months of engineering time by conventional methods.
Coding benchmarks show the clearest improvement over Opus 4.8. Fable 5 ranks first among frontier models on Cognition's FrontierCode evaluation at medium effort, reaches state-of-the-art performance on CursorBench, and posts the highest score Anthropic has measured on ViBench (a benchmark for unstructured "vibe-coding" sessions). On a proprietary spreadsheet-task evaluation, Fable 5 completes work 25–30% faster than Opus 4.8 while using fewer output tokens per task.
Vision and persistent-memory benchmarks both moved forward. Fable 5 extracts precise numerical values from scientific figures and rebuilds working web application source code from screenshots alone. Using file-based persistent memory, the model played Slay the Spire and reached the final act three times more frequently than Opus 4.8 — an evaluation Anthropic uses to measure whether a model can carry state coherently across very long contexts.
Claude Fable 5 vs Claude Opus 4.8 — performance, pricing, and when to use each
Fable 5 is the first AI model to cross 90% on Hex's core analytics benchmark. The benchmark measures multi-step data tasks that span hours of analytical work; Fable 5 posts a 10-point improvement over Opus 4.8. Hebbia's finance benchmark records the highest senior-level reasoning score the lab has measured. IMC's trading-analysis evaluation reports strong performance across factual lookup, root-cause analysis, and expected-value reasoning.
Fable 5 costs $10 per million input tokens and $50 per million output tokens. That is twice Claude Opus 4.8's standard API pricing of $5 input and $25 output. Through June 22, 2026, Fable 5 is included at no additional cost on Anthropic's Pro, Max, Team, and seat-based Enterprise plans on Claude.ai; after that date, metered API pricing applies. On Overchat AI, the model is included as part of the standard subscription with no Anthropic API key required.
The model selection question is straightforward. Fable 5 delivers measurably better results per dollar on long autonomous agent runs, multi-file refactors, and dense analytical work where Opus 4.8 has been producing partial answers — despite the higher token cost. For shorter tasks Opus 4.8 already handles cleanly, the cost-per-output favors Opus 4.8 or Sonnet. Fable 5 automatically falls back to Opus 4.8 on a narrow set of high-risk prompts, so model selection rarely requires manual switching mid-session.
Enterprise deployments and scientific research results
Early enterprise reports describe completion of work that previously required supervision. Stripe completed the 50-million-line Ruby migration described above. Cognition (Devin), Hebbia (financial document analysis), and IMC (trading research) each reported that Fable 5 finished extended task chains in a single pass with fewer abandoned attempts than previous Claude models. The consistent pattern is reduction in supervision overhead on long-horizon work rather than improvement on isolated single-turn queries.
Scientific research is one of the documented capability frontiers. Anthropic demonstrated Fable 5 constructing a solar-eclipse prediction from first-principles physics and autonomously playing Factorio as a strategy-planning evaluation. The restricted Mythos 5 sibling accelerated parts of protein design by approximately 10× in Anthropic's evaluations and generated viable candidate proteins for 9 of 14 drug-discovery targets, outperforming dedicated protein language models on gene-therapy applications.
How to access Claude Fable 5 on Overchat AI
Claude Fable 5 is available on Overchat AI under the Pro subscription. The model appears directly in the model dropdown, alongside the rest of Anthropic's lineup and other frontier models from OpenAI, Google, and xAI. Sign-in works with email, Google, or Apple — no Anthropic API key required, and no separate enterprise contract. The Pro plan covers all Fable 5 usage at a flat monthly rate, which is the practical difference from the metered $10/$50 per-million-token pricing on Anthropic's own API.
For work where Claude Opus 4.8 produces partial or inconsistent results, Claude Fable 5 is the appropriate model to test. This includes multi-step refactors, hours-long agent runs, complex research synthesis, and analytical tasks involving large or unstructured input. Shorter conversational work runs more cost-effectively on Opus 4.8 or Sonnet, and the model dropdown on Overchat AI makes the switch a single click.
.svg){kind=icon}
