Anthropic released Claude Fable 5 on June 9, 2026, making its most capable model generally available to the public for the first time. Fable 5 is a Mythos-class model shipped with safety guardrails in place, and for a two-week window it is free on Pro, Max, Team, and seat-based Enterprise plans. That puts the strongest Claude ever released in front of every paying creator at no extra cost, which changes the calculus for anyone choosing a default model. This deep dive compares Fable 5 with Claude Opus 4.8 and the restricted Mythos 5 variant, and lays out what creators should actually do before the free window closes on June 22.
Background
Fable 5 is the public face of a model family Anthropic had kept behind Project Glasswing, its trusted-access program for security and research partners. The launch landed days after the company warned that frontier AI is getting more dangerous, a sequence that drew most of the press attention. For working creators, the more useful signal is that the public can use it today rather than the safety debate around it.
Availability is broad from day one. Fable 5 is live in the Claude apps, in the API under the identifier claude-fable-5, generally available inside GitHub Copilot, and already shipped in vertical tools such as legal-AI platform Harvey. That matters because a model creators can only reach through one app rarely changes a workflow, while one that spans chat, API, and coding assistants on launch day does.
Deep Analysis
What "Mythos-class, made public" actually means
Anthropic separates raw capability from the safeguards wrapped around it. Mythos 5 is the unrestricted base model; Fable 5 is the same base model with safety classifiers attached, sold to everyone. Until now the Mythos tier only reached vetted partners, so the headline here is access, not a brand-new architecture. Anthropic describes Fable 5 as state-of-the-art on nearly all tested benchmarks of AI capability, and points to specific external results: top score among frontier models on Cognition's FrontierCode, the highest score any model has posted on Hebbia's finance benchmark, state-of-the-art on CursorBench, and the first model to break 90 percent on a core analytics benchmark. The company did not publish a context-window figure or a single headline numeric score, so the honest read is "best-in-class on the evaluations vendors run," not a clean one-number win.
Fable 5 vs Opus 4.8 vs Mythos 5: capability and access
The three models are best understood as one capability tier sliced by access and safety posture. Fable 5 is the strongest model a creator can actually buy. Mythos 5 is the same engine without guardrails, locked to Glasswing cyberdefense partners and select biology researchers. Opus 4.8, the prior flagship, now does double duty: it is still a strong general model and it is the safety fallback for Fable 5. The table below maps how they differ on the dimensions that affect a creator's choice.
| Dimension | Claude Fable 5 | Claude Mythos 5 | Claude Opus 4.8 |
|---|---|---|---|
| Who can use it | Public: API, Claude apps, Copilot, partners | Glasswing partners and select researchers only | Public |
| Capability tier | Mythos-class, current SOTA on most vendor benchmarks | Same base as Fable 5, no safeguards | Prior flagship, strong general model |
| Safety posture | Classifiers on; high-risk queries fall back to Opus 4.8 | Safeguards lifted | Acts as the Fable 5 fallback target |
| API price (per million tokens) | $10 input / $50 output | Restricted access, not sold per-seat | Anthropic standard tier |
| Best for creators | Long edits, screenshot-to-code, multi-step agentic jobs | Not creator-facing | Reliable default once the free window closes |
One customer datapoint from Anthropic's announcement gives a sense of the gap at the top: on a physics research task, Fable 5 reportedly delivered in 36 hours what GPT-5.5 took four days to produce. Treat that as a single anecdote, not a benchmark, but it points the same direction as the FrontierCode and CursorBench results.
The safeguard tax and the Opus 4.8 fallback
Fable 5 routes around its own guardrails rather than refusing outright. Three classifier categories trigger an automatic fall back to Opus 4.8: cybersecurity queries, most biology-and-chemistry requests, and prompts flagged as model-distillation attempts. Anthropic says this fallback fires in fewer than 5 percent of sessions, and that Mythos 5's measured rate of misaligned behavior was low and similar to Opus 4.8. For most creative work, image briefs, scripts, code, research, none of this is in play. The edge case to know about is technical or scientific content that brushes the bio-chem or security classifiers: in those moments you are quietly served Opus 4.8 instead of Fable 5, so a sudden dip in output quality on a niche prompt is the system working as designed, not a bug.
The pricing math: free window, then usage credits
Through the API, Fable 5 costs $10 per million input tokens and $50 per million output tokens, which Anthropic says is less than half the price of the earlier Mythos Preview. The headline for subscribers is the window: from June 9 through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost. From June 23 onward it draws down usage credits on those plans, and consumption-based Enterprise and API customers pay the per-token rate from day one. Exact rate-card details live in the Claude API documentation. The practical implication: the next two weeks are a free, full-strength trial, and the cost question only becomes real on June 23.
Impact on Creators
The reason to care is concrete: for two weeks, the most capable model Anthropic sells is free to anyone already on a paid Claude plan, and it holds focus across long, multi-step work better than any prior Claude. That is exactly the profile that long-form drafting, full-codebase creative tooling, and book-length editing passes need. The free window is a clean opportunity to benchmark Fable 5 against your current default on real projects, not demos, before deciding whether the extra capability justifies usage credits once billing kicks in. Anthropic's broader Mythos push, including scaling Claude Mythos to critical infrastructure, signals this tier is where the company is investing, and it arrives the same month Apple expanded its own on-device foundation models, sharpening the cloud-versus-device choice creators now make per task.
Key Takeaways
- Fable 5 is the public, safeguarded version of Anthropic's Mythos-class model, generally available June 9, 2026, across API, Claude apps, GitHub Copilot, and partner tools.
- It is free on Pro, Max, Team, and seat-based Enterprise plans from June 9 to 22, then moves to usage credits on June 23.
- API pricing is $10 input and $50 output per million tokens, described as less than half the price of Mythos Preview.
- High-risk cyber, bio-chem, and distillation queries silently fall back to Opus 4.8 in fewer than 5 percent of sessions.
- Mythos 5, the unguarded base model, stays restricted to Glasswing partners and select researchers and is not creator-facing.
Frequently Asked Questions
How is Claude Fable 5 different from Claude Mythos 5?
They share the same base model. Fable 5 ships with safety classifiers and is sold to everyone; Mythos 5 has those safeguards lifted and is restricted to Project Glasswing cyberdefense partners and select biology researchers.
Is Claude Fable 5 actually free?
Yes, but only on a window. From June 9 through June 22, 2026, it is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. After June 22 it draws down usage credits, and API or consumption-based Enterprise customers pay the per-token rate from the start.
How does Fable 5 compare to Claude Opus 4.8?
Fable 5 is the newer, more capable flagship and posts state-of-the-art results on most vendor benchmarks. Opus 4.8 remains a strong general model and now also serves as the safety fallback that Fable 5 routes high-risk queries to.
Why would Fable 5 sometimes give weaker answers?
Three classifier categories, cybersecurity, most biology and chemistry, and suspected distillation, trigger an automatic fall back to Opus 4.8. This fires in under 5 percent of sessions, so a quality dip on a niche technical prompt usually means you were served the fallback model.
Where can I use Claude Fable 5?
In the Claude apps, in the API under the identifier claude-fable-5, inside GitHub Copilot where it went generally available, and in partner products such as Harvey.
What to Watch
The first thing to track is what happens on June 23, when the free window closes and usage credits start metering Fable 5 on subscription plans. That is when creators learn whether the model earns a permanent slot or stays a special-occasion tool, and whether Anthropic adjusts plan allowances in response to demand. The second is independent benchmarking: the launch leaned on vendor evaluations like FrontierCode, CursorBench, and Hebbia's finance test, so the more revealing numbers will come from third parties running Fable 5 against GPT-5.5 and Gemini on reproducible tasks. The third is the safeguard boundary, since the Opus 4.8 fallback is the clearest sign yet that Anthropic is shipping capability and containment as a single product, and where that line sits will shape how much of the frontier any creator actually gets to use.
