Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, the first major update to its Mythos-class model family after the messy Fable 5 rollout earlier this summer.
The short version: Fable 5.1 is generally available, Mythos 5.1 remains restricted, and Anthropic says the new release is stronger on long-running coding, knowledge work, agentic tasks, and scientific research. The commercial hook is just as important: Anthropic says the model should be cheaper than Fable 5 in real agent workflows because cache reads now cost much less.
This is not a simple model refresh. It is Anthropic trying to fix three objections that followed Fable 5: price, data retention, and safeguards that sometimes blocked benign work.
What changed
Claude Fable 5.1 is positioned as Anthropic's most capable generally available Claude model for demanding work. The platform docs list the model ID as claude-fable-5-1, with a 1 million token context window, 128,000 token maximum output, adaptive thinking, and a June 2026 knowledge cutoff.
The input and output prices stay at the Fable tier:
| Item | Price |
|---|---|
| Input tokens | $10 per 1M tokens |
| Output tokens | $50 per 1M tokens |
| 5-minute cache write | $12.50 per 1M tokens |
| 1-hour cache write | $20 per 1M tokens |
| Cache read | $0.25 per 1M tokens |
The important change is the cache-read price. Anthropic says cache reads are now 75% cheaper, which should cut typical billed workloads by about 25% versus Fable 5. For complex coding and highly agentic work, Anthropic says the savings can reach roughly 45%.
That matters because agentic systems often reuse long context: repository maps, instructions, file summaries, tool traces, plans, test results, and prior decisions. If the model can keep reading cached context cheaply, the per-task cost can fall even when the headline input/output price remains high.
Fable 5.1 versus Mythos 5.1
Anthropic says Fable 5.1 and Mythos 5.1 are the same underlying model with different safeguards.
Fable 5.1 is the version most users can access through Claude.ai, Claude Platform, AWS, Google Cloud, and Microsoft Azure. Mythos 5.1 is available only through trusted access programs for vetted work in cybersecurity and life sciences.
That split continues the design Anthropic introduced with Claude Fable 5 and Mythos 5: public access to the high-end model family, but with restricted handling for domains that Anthropic treats as high-risk. The difference this time is that Anthropic says the safeguards are more precise.
In cybersecurity, Fable 5.1 can now be used to identify software vulnerabilities for defensive work, but still routes exploit development, penetration testing, and binary-focused vulnerability scanning away from Fable. Anthropic says Claude Code users should see about 60% fewer cyber-safeguard interventions per session than they saw with the previous Fable 5 safeguards.
In biology, Anthropic says benign elementary biology and medical requests should trigger safeguards far less often than the original Fable 5 setup, while research-and-development work in life sciences remains limited to approved Mythos access.
Benchmark claims
Anthropic's launch materials put Fable 5.1 ahead of Fable 5 across several agentic and knowledge-work benchmarks, with the biggest story being performance per cost rather than a raw-score sweep.
Key numbers from Anthropic's benchmark table include:
| Benchmark | Fable 5.1 | Fable 5 | Opus 5 |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | 29.0% |
| Terminal-Bench 4.0 | 55.8% | 42.0% | 52.3% |
| GDPval-AA v2 | 1853 | 1723 | 1824 |
| OSWorld 2.0 partial | 77.9% | 72.9% | 75.4% |
| OSWorld 2.0 strict | 41.7% | 36.1% | 39.6% |
| Humanity's Last Exam, no tools | 60.9% | 57.8% | 56.6% |
| Humanity's Last Exam, with tools | 65.0% | 63.8% | 63.6% |
| AutomationBench | 31.4% | 17.1% | 26.9% |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% |
The caveat is important: these are Anthropic's own release benchmarks, and the company notes that production safeguards can affect some results. Still, the pattern fits the product pitch. Fable 5.1 is aimed at long-horizon work where the model plans, uses tools, verifies its own output, and continues over many steps.
That puts it directly in the path of professional tools like Cursor, Claude Code, and enterprise agent systems. For software teams, the question is not only whether Fable 5.1 scores higher, but whether it completes more real tasks per dollar.
Enterprise data controls
Anthropic is also using this launch to introduce Enterprise Frontier Safeguards, or EFS.
The promise is narrow but important: eligible enterprise customers should be able to use Covered Models like Fable 5.1 with privacy similar to zero data retention while still preserving safety monitoring. Anthropic says EFS works by keeping data in customer-controlled cloud infrastructure, with human review handled by the customer by default rather than Anthropic.
EFS is scheduled to roll out in phases beginning later this fall. Until then, Anthropic says eligible customers can use Fable 5.1 with zero data retention.
That is a meaningful shift for regulated teams. Fable 5 previously carried a 30-day retention default for safety monitoring, which made some enterprise use harder to approve. EFS looks like Anthropic's attempt to keep high-risk-model monitoring without forcing customers to accept Anthropic-side data storage.
Why this matters
Fable 5.1 is a response to a very specific market problem: frontier models are getting more useful for serious work, but harder to deploy cleanly.
The earlier Fable 5 access disruption showed how quickly a high-end model can become an operational dependency. The later access extension showed that pricing and entitlement rules were still unsettled. Fable 5.1 tries to make the model easier to adopt by reducing cache-heavy task cost, reducing false positives, and giving enterprises a clearer data-control path.
For developers and AI teams, the practical takeaway is simple: Fable 5.1 is probably not the default model for every prompt. Anthropic's own docs still suggest starting most work with Opus 5 and moving to Fable 5.1 when demanding reasoning or long-horizon agentic work justifies it.
That makes Fable 5.1 a specialist model: expensive on paper, potentially cheaper per completed task when the work is long, cached, tool-heavy, and difficult.
Our take
Claude Fable 5.1 looks like Anthropic's cleanest version of the Mythos-class bet so far.
The model story is stronger coding and research performance. The business story is cheaper agent loops through cache reads. The enterprise story is EFS. The safety story is more precise restrictions rather than simply heavier ones.
The open question is whether the real-world economics match Anthropic's examples. If Fable 5.1 genuinely completes long coding and knowledge-work tasks with fewer retries, fewer safeguard interruptions, and lower cached-context cost, it becomes easier to justify than Fable 5. If not, it remains a premium model teams reserve for the hardest workflows.
Either way, this is the right direction for Anthropic. The frontier-model race is no longer only about who tops a benchmark table. It is about who can package capability, cost, safety, and enterprise controls into something teams can actually run.