Claude Fable 5 Explained: Pricing, Features & How It Compares to Opus, Sonnet and Haiku (2026)
Anthropic has launched Claude Fable 5, the first model in its new Claude 5 family — and the first in a brand-new "Mythos-class" tier that sits above Claude Opus in capability. If you've been wondering what Claude Fable 5 actually is, how much it costs, and whether you should switch from Opus, Sonnet, or Haiku, this guide covers everything.
What Is Claude Fable 5?#
Claude Fable 5 (model ID: claude-fable-5) is Anthropic's most intelligent generally available AI model. It's designed for the most demanding reasoning and long-horizon agentic work — think overnight autonomous coding runs, complex multi-step research, end-to-end enterprise deliverables like financial models and full document workflows.
A few headline specs:
- 1M token context window (the maximum is also the default)
- 128K max output tokens per request
- Thinking is always on — the model automatically decides how deeply to reason on every request
- Best-in-class long-horizon agentic execution — single requests on hard tasks can run for many minutes as the model gathers context, builds, and verifies its own work
Unlike previous Claude releases that were incremental version bumps (Opus 4.6 → 4.7 → 4.8), Fable 5 introduces a new naming scheme and a new capability tier entirely.
Claude Fable 5 vs Claude Mythos 5: What's the Difference?#
Alongside Fable 5, Anthropic introduced Claude Mythos 5 (claude-mythos-5). Here's the key thing to understand: they are the same underlying model.
- Claude Fable 5 is generally available and includes additional safety measures for dual-use capabilities (it runs safety classifiers targeting research biology and most cybersecurity content).
- Claude Mythos 5 offers the same capabilities and pricing but without those measures — and it's available only to approved organizations through Anthropic's Project Glasswing program.
For virtually everyone, Fable 5 is the model you'll actually use. Mythos 5 succeeds the earlier invitation-only Claude Mythos Preview.
Claude Fable 5 Pricing#
Claude Fable 5 is priced above the Opus tier, reflecting its position as the new flagship:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude Fable 5 | $10.00 | $50.00 | 1M |
| Claude Opus 4.8 | $5.00 | $25.00 | 1M |
| Claude Sonnet 4.6 | $3.00 | $15.00 | 1M |
| Claude Haiku 4.5 | $1.00 | $5.00 | 200K |
That's exactly 2× the price of Opus 4.8 on both input and output. Prompt caching (cache reads at roughly one-tenth the input price) and batch processing can significantly reduce effective costs for production workloads.
Claude Fable 5 vs Opus 4.8: Should You Upgrade?#
Opus 4.8 remains Anthropic's recommended default for most work. Fable 5 is worth the premium when:
- Your tasks are at the top of your difficulty range. The teams with the best early results gave Fable 5 their hardest unsolved problems — large migrations, deep research, first-shot implementations of well-specified systems.
- You run long autonomous agents. Fable 5 sustains parallel sub-agent delegation and multi-hour runs more reliably than any prior model.
- You need the strongest vision on messy inputs. It's explicitly trained to handle flipped, blurry, or noisy images using cropping and processing tools.
Stick with Opus 4.8 when your workloads are already handled well — Fable 5's biggest gains are on work above what prior models could do, not on tasks Opus already solves.
Key API Differences#
If you're a developer, Fable 5 behaves differently from Opus in a few important ways:
- Thinking is always on. You can't disable it — sending
thinking: {type: "disabled"}returns an error. Depth is controlled with theeffortparameter (lowthroughmax). - The raw chain of thought is never returned. You can request a readable summary of the reasoning instead.
- Safety refusals are possible. A request may return a
refusalstop reason; Anthropic recommends configuring an automatic fallback to Opus 4.8 so declined requests are transparently re-served. - No assistant prefill, same as the rest of the Claude 4.6+ family.
- 30-day data retention required — Fable 5 isn't available under zero-data-retention configurations.
Fable 5 vs Sonnet 4.6 vs Haiku 4.5: Which Should You Choose?#
| Use case | Best model | Why |
|---|---|---|
| Hardest reasoning, long autonomous agents | Fable 5 | Highest intelligence ceiling, minutes-long deep runs |
| Most coding and knowledge work | Opus 4.8 | Excellent capability at half the price |
| High-volume production, chat apps | Sonnet 4.6 | Best speed/intelligence balance |
| Classification, simple tasks, speed-critical | Haiku 4.5 | Fastest and cheapest |
A practical pattern many teams use: run the orchestrator on Fable 5 or Opus 4.8, and delegate simple sub-tasks to Sonnet or Haiku sub-agents to keep costs down.
How to Access Claude Fable 5#
- Claude API — use model ID
claude-fable-5(available in all official Anthropic SDKs) - Claude apps — available in Claude's web, desktop, and mobile apps on paid plans
- Claude Code — select it as your model for the hardest agentic coding work
- Cloud platforms — availability on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry follows each platform's rollout schedule
Frequently Asked Questions#
Is Claude Fable 5 better than GPT models?#
Fable 5 is Anthropic's strongest model to date, with particular advantages in long-horizon agentic work, code review and debugging, and sustained multi-agent coordination. Head-to-head comparisons depend on the task — see our ChatGPT vs Claude comparison for the broader picture.
Why is it called "Fable" instead of "Opus 5"?#
Anthropic introduced a new Mythos-class tier above Opus. Fable 5 is the generally available member of that tier; Mythos 5 is the unrestricted variant for approved organizations. Anthropic's announcement is at anthropic.com/news/claude-fable-5-mythos-5.
Does Claude Fable 5 replace Opus?#
No. Opus 4.8 remains actively developed and is still the recommended default for most workloads. Fable 5 sits above it as a premium tier.
What is the context window of Claude Fable 5?#
1 million tokens — and unlike some models where long context is opt-in, 1M is the default.
The Bottom Line#
Claude Fable 5 is the new ceiling for AI capability in 2026: a genuinely more intelligent model tier, priced accordingly at $10/$50 per million tokens. If your work already pushes Opus 4.8 to its limits — long autonomous agents, hard reasoning, complex enterprise deliverables — Fable 5 is worth testing on your hardest problems first. For everything else, Opus 4.8 and Sonnet 4.6 remain the smarter buys.
Want to compare AI models and tools side by side? Explore our tool directory or check out the best AI coding tools for 2026.
