Claude Fable 5 Free for Two Weeks: 80.3% SWE-Bench, Where to Use It

Anthropic is giving away its most capable model for two weeks. Claude Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost from June 9 through June 22, 2026, after which it moves to usage credits. The free window matters because the model posted 80.3% on SWE-Bench Pro, the agentic coding benchmark where GPT-5.5 scores 58.6%.

We covered what shipped and what stays gated on launch day. This is the day-two guide: where Fable 5 is actually available, what the fine print says, and how to use the free window before June 23.

Where Claude Fable 5 Is Available Today

Fable 5 went live on every major developer surface within 24 hours of launch, which has never happened for an Anthropic frontier model before.

The Copilot listing covers VS Code, Visual Studio, the Copilot CLI, the cloud agent, github.com, GitHub Mobile, JetBrains, Xcode, and Eclipse. If you do not see the model yet, the rollout is gradual rather than instant.

The Free Window: June 9 to June 22

Anthropic is waiving usage credits for Fable 5 on subscription plans for exactly two weeks, and that is a deliberate evaluation funnel, not generosity. After June 23, Fable 5 calls on those same plans consume usage credits, and API traffic bills at list price from day one.

The play is obvious once you see the pricing ladder: get developers hooked on the strongest model during the free period, then let the workloads that genuinely need it graduate to paid usage. It mirrors what GitHub Copilot did with token billing, except Anthropic is running the trial at its own expense first.

Benchmarks: 80.3% SWE-Bench Pro, and the Gap Is the Story

On agentic coding, the distance between Fable 5 and everything else is larger than any frontier release this year.

Fable 5 also posted 85.0% on OSWorld-Verified (computer use) and edged Opus 4.8 on knowledge work, scoring 1932 against 1890 on GDPval-AA. Anthropic's repeated framing held up in third-party summaries of the system card: the longer and more complex the task, the larger the lead. That matches the launch-day proof point of Stripe migrating a 50-million-line Ruby codebase in a day.

For the full four-model picture including subscription pricing, see our updated Claude vs ChatGPT vs Gemini vs Grok comparison and the best AI models 2026 hub.

The Benchmark Asterisk Most Coverage Missed

Several headline scores in the system card belong to Mythos 5, not the Fable 5 you can actually buy. The starred cybersecurity and biology numbers were measured on the unrestricted Mythos variant. On offensive cyber tasks, Mythos 5 scored 78.0% on ExploitBench while Fable 5 in blocking mode made 0% progress, because the safety classifiers hand those queries to Opus 4.8.

That is not a criticism, it is the product working as designed. But it means two things for anyone evaluating the model. First, if your workload touches security tooling, pentest automation, or bioinformatics, benchmark Fable 5 yourself rather than trusting the card, since you will hit the fallback path. Second, the fallback fires in under 5% of sessions overall, with a notice, so for ordinary product and coding work the published coding and reasoning numbers are the real ones.

API Pricing and the Caching Math

Fable 5 costs exactly double Opus 4.8: $10 per million input tokens and $50 per million output, against $5/$25 for Opus 4.8. The number that changes the economics is the 90% prompt caching discount on input tokens.

Agentic coding workloads resend large, stable contexts on every turn: system prompts, repo maps, tool definitions. With caching, that repeated input bills at roughly $1 per million instead of $10, which puts long-running Fable 5 agent sessions closer to uncached Opus pricing than the sticker numbers suggest. Run your own traffic shape through our LLM API pricing tracker before deciding the model is too expensive, and if you are still picking a daily driver, the Claude vs ChatGPT quiz takes two minutes.

The GitHub Copilot Fine Print

Two administrative details will trip up enterprise teams. The Fable 5 policy in Copilot is disabled by default for Business and Enterprise, so an admin has to switch it on before anyone sees the model. And enabling it requires acknowledging Anthropic's 30-day data retention requirement for safety classifiers, a condition no other Claude model in Copilot carries.

That retention clause exists because the classifier pipeline that routes risky queries to Opus 4.8 needs logs to audit. Compliance teams at banks and healthcare companies will want to read it before flipping the switch, and that review cycle, not the technology, is what will pace enterprise adoption this month.

Our Analysis: How to Run the Two-Week Evaluation

The free window is twelve more days as of this writing. A concrete plan for getting a real answer out of it:

1. Pick one long-horizon task, not twenty prompts. Fable 5's lead concentrates in multi-hour autonomous work. A real migration, a cross-repo refactor, or a test-suite overhaul will tell you more than any chat session.
2. Log the fallback notices. If Opus 4.8 fires on more than a handful of your sessions, your domain leans on the restricted categories and your effective model is partly Opus. Price and plan accordingly.
3. Measure tokens per completed task, not per request. Anthropic claims Fable 5 finishes work in fewer turns with less reasoning overhead. If that holds on your workload, the 2x unit price can net out cheaper than Opus 4.8 per outcome.
4. Decide before June 23. After the window closes, the same evaluation costs real usage credits. Teams that wait will end up benchmarking on the meter.

Key Takeaways

• June 9 to June 22, 2026: Claude Fable 5 is included free on Pro, Max, Team, and Enterprise plans; usage credits apply from June 23
• 80.3% SWE-Bench Pro against GPT-5.5's 58.6% and Gemini 3.1 Pro's 54.2%, with the lead growing on longer tasks
• $10/$50 per million tokens, double Opus 4.8, but a 90% prompt caching discount reshapes the math for agentic workloads
• The asterisk: starred cyber and bio benchmark scores belong to gated Mythos 5; Fable 5 routes those queries to Opus 4.8 and scored 0% on offensive cyber in blocking mode
• Copilot fine print: model policy is off by default for Business/Enterprise and requires acknowledging a 30-day data retention term
• For developers: spend the free window on one real long-horizon task, log fallback notices, and measure cost per completed task before June 23
• What to watch: whether Anthropic extends the free window, and the first independent SWE-Bench Pro replications expected within two weeks

Sources

• Claude Fable 5 is generally available for GitHub Copilot — GitHub Changelog, June 9 2026
• Anthropic brings Mythos to the masses with Claude Fable 5 — VentureBeat, June 9 2026
• Claude Fable 5 from Anthropic now available on Amazon Bedrock — About Amazon, June 9 2026
• Claude Fable 5 and Mythos 5: release, benchmarks — Digital Applied, June 2026