Claude Fable 5: Anthropic's most powerful model

Claude Fable 5 is Anthropic's most capable generally-available model — released June 9, 2026 as a new tier above Opus, not an Opus upgrade. It ships a 1M-token context window with a 128K-token output limit (the largest output budget in the Claude line), state-of-the-art reasoning, and training aimed squarely at long-horizon agentic work: tasks that run for hours across many tool calls without losing the plan. It uses adaptive thinking only — the model decides how much to reason per request, and there are no sampling parameters to tune. Pricing is $10 per million input tokens and $50 per million output: double Claude Opus 4.8 ($5/$25) and the most expensive model in Council's July 2026 lineup. The case for Fable 5: research-grade reasoning, hard multi-step agent runs, and any task where you'd otherwise chain multiple Opus calls. The case against: everyday coding and writing, where Opus 4.8 or Sonnet 5 deliver most of the quality at half to a fifth of the price. Rule of thumb — Fable 5 is the escalation model, not the default.

Specs

Best at

• Long-horizon agentic work — sustains multi-hour, many-tool-call runs without drifting off plan.
• State-of-the-art reasoning — Anthropic's top score on hard reasoning workloads; the escalation target when Opus 4.8 stalls.
• Very long outputs — the 128K output limit fits full reports, large refactor patches, and book-chapter drafts in one response.
• Research synthesis — 1M input + 128K output means whole-corpus in, structured monograph out.

Where it loses

• Price — $10/$50 is 2× Opus 4.8 and over 3× Sonnet 5. Most daily tasks don't need it.
• No sampling control — adaptive thinking only; you can't set temperature or force a fast shallow pass.
• Latency on simple prompts — a frontier reasoner is overkill for classification, chat, and lookup work.

Fable 5 vs Opus 4.8: a new tier, not a version bump

Before June 2026, Opus was the top of the Claude range. Fable 5 changes the ladder: it sits above Opus 4.8 rather than replacing it. Both run 1M context, vision, and native web search — the deltas are reasoning ceiling, the 128K output limit, and price. Opus 4.8 ($5/$25) remains Anthropic's coding and agent workhorse; Fable 5 ($10/$50) is what you reach for when the task is genuinely hard: ambiguous research questions, agent runs that keep failing partway, architecture decisions with long dependency chains. If you're unsure which you need, you probably need Opus 4.8.

How Council AI runs it

Fable 5 is available on every paid Council AI plan (Plus, Pro, and Ultra) with no separate Anthropic account or API key. In a council run it answers in parallel with models from up to 8 other labs — OpenAI, Google, xAI, DeepSeek, Qwen, Mistral, Moonshot, Z.ai — and the moderator synthesizes a consensus answer with an agreement score. Because Fable 5 is the priciest voice in the lineup, a common pattern is to seat it in councils only for high-stakes prompts and let cheaper models carry routine fan-outs; your plan's monthly budget cap keeps the spend bounded either way.

Frequently asked questions