Anthropic launched Claude Fable 5.1 (generally available) and Claude Mythos 5.1 (trusted access only). The company says they are the same underlying model with different safeguard levels. Fable 5.1 is priced for roughly 25% lower typical-workload cost via cheaper cache reads (company estimate; higher savings claimed on highly agentic work). Cyber discovery is allowed on Fable; exploit development is not. Mythos is for vetted cyber and life-sciences programs. Enterprise Frontier Safeguards / zero-data-retention privacy is a phased enterprise claim—not a same-day universal switch.
Introducing Claude Fable 5.1 · Anthropic
Quick Take
Anthropic’s September 2026 launch splits one capability story into two products: Claude Fable 5.1 is generally available; Claude Mythos 5.1 uses the same underlying weights with different safeguards and is limited to trusted access programs. The practical news for builders is price shape (cheaper cache reads), sharper cyber permissions on Fable (discovery yes, exploit development no), and a clear statement that Mythos is where vetted cyber and life-sciences work is supposed to live—not an open public SKU.
What the video shows
Anthropic’s official “Introducing Claude Fable 5.1” video (ROF2Nv_KjOM) is the clean embed for this brief: company framing of the GA model. It will not, by itself, audit Mythos enrollment criteria or reproduce benchmark tables—read the product post and system card for those.
What is new
Anthropic states Fable 5.1 and Mythos 5.1 are the same model with different levels of safeguards. Fable 5.1 ships across Anthropic’s platforms (API id claude-fable-5-1 called out in the post) and cloud partners. Mythos 5.1 is available only through trusted access programs aimed at cybersecurity and life sciences.
Pricing (company claim): Anthropic estimates Fable 5.1 will cost about 25% less than Fable 5 for typical workloads wherever usage is billed by token, driven by a large cut to cache-read pricing (Anthropic says cache reads now cost 75% less, or $0.25 per million tokens). For highly agentic work, Anthropic says savings can reach roughly 45%. Base input/output token prices are otherwise described as matching Fable 5 ($10 / $50 per million).
Safeguards and cyber posture (company claim): Fable 5.1 may be used to discover software vulnerabilities, but not to develop exploits. Anthropic says cyber safeguards produce about 60% fewer false-positive interventions per Claude Code session versus prior Fable 5 safeguards, while still redirecting some dual-use tasks (including penetration testing, exploit generation, and certain binary scanning) to other models. Mythos pairs more permissive safeguards for vetted users via the Cyber Verification Program (CVP) and Life Sciences Verification Program (LSVP), developed in partnership with the U.S. government for the biology track. Claude Security is also described as powered by Mythos 5.1.
Privacy / EFS (phased): Enterprise Frontier Safeguards are pitched as customer-controlled cloud storage with privacy comparable to zero data retention while retaining misuse detection. Anthropic says EFS rolls out in phases beginning later this fall; until then, eligible customers can use Fable 5.1 with zero data retention. That is an eligibility-and-timeline claim, not ‘everyone gets ZDR today.’
Benchmarks and science demos: Anthropic publishes large internal/partner benchmark tables (Terminal-Bench, CursorBench, HLE, OSWorld, etc.) and research vignettes (protein binders under Mythos, Venus DEM work, GPU kernel speedups). Treat these as company-reported results pending independent reproduction.
What evidence supports
1. Primary product post — Anthropic’s Fable/Mythos 5.1 announcement defines GA vs trusted access, same-weights framing, price and safeguard changes, and program names (CVP/LSVP).
2. Explicit cyber split — Discovery allowed on Fable; exploit development disallowed; dual-use tasks redirected; Mythos for vetted defensive/research contexts per company policy text.
3. Cache-read price mechanics — Anthropic’s stated $0.25/M cache-read price and estimated 25%/45% workload savings give operators a concrete (still company-sourced) cost hypothesis to test on their own traces.
4. EFS timeline language — Phased fall rollout + interim ZDR for eligible customers is written as future/conditional, which is the correct buying-process expectation.
5. Official video — Anthropic-labeled intro clip matches the GA product narrative for embed purposes.
What it does not prove
- That Mythos 5.1 is open to every developer who wants fewer refusals.
- That Anthropic’s benchmark leads will hold under third-party harnesses identical to yours.
- That ~25% / ~45% savings will appear on every invoice mix (cache-read share varies).
- That EFS is already live as a universal enterprise default.
- That allowing vulnerability discovery on Fable means Anthropic endorses offensive exploit pipelines—company text says the opposite for exploit development.
- That scientific vignettes (binders, Venus map, kernels) generalize to your lab’s reliability bar without local validation.
Why this matters
For operators, the actionable story is SKU literacy. If your workload is coding agents and knowledge work, Fable 5.1’s GA status plus cache-read economics is the near-term evaluation target—measure refusal rates on defensive vuln research separately from any fantasy of Mythos-level permissions. If you need Mythos-class cyber or life-sciences latitude, budget time for verification programs rather than assuming the marketing headline equals access. Same weights with different gates is how Anthropic is trying to sell capability without a single open frontier surface; buyers should underwrite the gate, not only the model card.
Sources
- Anthropic — Introducing Claude Fable 5.1 and Claude Mythos 5.1: https://www.anthropic.com/claude-fable-and-mythos-5-1
- Anthropic system card (referenced in product post; review at publish time for eval detail)
- Video: Anthropic — Introducing Claude Fable 5.1 — https://www.youtube.com/watch?v=ROF2Nv_KjOM