Anthropic Releases Claude Fable 5.2 and Mythos 5.2 With Faster Output and Lower Prices
Anthropic's updated frontier models generate output up to 30% faster, list at $8 per million input tokens and $40 per million output tokens, and are built to orchestrate teams of smaller Claude models.
By The AI Enterprise Magazine Team5 min readNews · Weekly briefing
Key takeaways
- Faster and cheaper. Fable 5.2 and Mythos 5.2 produce output up to 30% faster than 5.1 and list at $8 per million input and $40 per million output tokens.
- Available everywhere. Fable 5.2 launched on Anthropic's platform plus AWS, Google Cloud and Microsoft Azure; Mythos stays limited to vetted cyber and life sciences groups.
- Built to orchestrate. Fable can plan and delegate to Haiku, Sonnet or Opus, which favors tiered agent designs that cut cost per task.
- Vendor numbers. Benchmark gains such as 49.1% on FrontierCode 2.0 are Anthropic's own results and need internal testing.
Week of November 9, 2026
Anthropic's 5.2 Models Post Their Biggest Gains on Coding and Robotics Tests
Fable 5.2 more than doubled its robotics simulation score and lifted its FrontierCode 2.0 result by about 17 points over Fable 5.1, while Mythos 5.2 posted smaller gains on security and biology tests.
- 5.1 model
- 5.2 model
View the data
| 5.1 model | 5.2 model | |
|---|---|---|
| Fable: RoboDojo Sim | 27.6% | 64.9% |
| Fable: FrontierCode 2.0 | 32.4% | 49.1% |
| Mythos: CyScenarioBench | 61.7% | 76.3% |
| Mythos: CVE Rediscovery | 78.3% | 89.6% |
| Mythos: 24-hour binder design | 79.1% | 85.1% |
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9, an update to its two most capable model families that the company says produces output up to 30% faster than the 5.1 versions and arrives at a lower list price. Fable 5.2 is open to all customers and is already available across Anthropic's own platform and the three largest clouds, while Mythos 5.2 remains limited to vetted cybersecurity and life sciences organizations. For enterprise data and AI leaders, the release matters less as a leaderboard event than as a change to the cost and routing decisions behind agent workloads that already run on Anthropic models.
What did Anthropic release?
Fable 5.2 is Anthropic's general-purpose frontier model for long, open-ended technical work, and Mythos 5.2 is the restricted variant tuned for security and biology research. Both generate output up to 30% faster than their 5.1 predecessors, according to SD Times and 9to5Mac, which quoted Anthropic describing Fable as the model "best for the most ambitious, open-ended technical work." An optional fast mode raises token generation speed by up to 2.5 times, TestingCatalog reported.
Anthropic also published a system card alongside the models, Let's Data Science reported. The company said Fable's built-in cyber and biology safeguards are what allow it to be offered broadly, and TestingCatalog reported that those safeguards produced about 55% fewer blocks than Fable 5.1 in early testing, which should mean fewer false refusals on legitimate security and engineering work.
How much does Fable 5.2 cost, and where can enterprises get it?
Standard pricing for Fable 5.2 is $8 per million input tokens and $40 per million output tokens, according to all four outlets that covered the launch. 9to5Mac quoted Anthropic applying the same rate to Mythos 5.2. Fast mode costs twice as much, at $16 per million input tokens and $80 per million output tokens, TestingCatalog and Let's Data Science reported. Anthropic described the new rate as lower than before, but none of the coverage we reviewed listed the 5.1 prices, so the size of the cut is not yet clear.
Availability is broad from day one. SD Times reported that Fable 5.2 "is available today on all platforms," and TestingCatalog said it can be used through Claude's own products and API as well as Amazon Web Services, Google Cloud and Microsoft Azure. Inside Anthropic's subscription plans, 9to5Mac reported that Fable can account for up to 50% of weekly usage limits on Pro plans and standard Team seats. Mythos 5.2 is offered only to organizations approved through Anthropic's Life Sciences and Cyber Verification Programs, according to SD Times.
How does Fable 5.2 perform on Anthropic's benchmarks?
The headline numbers are Anthropic's own, so they should be read as vendor claims until independent testing catches up. On FrontierCode 2.0, a coding benchmark run at the highest effort setting, Fable 5.2 scored 49.1% against 32.4% for Fable 5.1 and 45.5% for Claude Opus 5.5, Let's Data Science reported. On FrontierSWE v2 it scored 69.9% against 65.5% for OpenAI's GPT-6 Astra. On an internal robotics simulation test called RoboDojo Sim, the score rose to 64.9% from 27.6%.
The results are not a clean sweep. Google's Gemini 4 Carbon led Fable 5.2 on AutomationBench, 53.9% to 51.4%, and Opus 5.5 edged it on the AA-Briefcase Elo ranking, according to the same report. On the restricted model, Mythos 5.2 rediscovered 89.6% of known vulnerabilities in a CVE rediscovery test, up from 78.3% for Mythos 5.1, and scored 76.3% on CyScenarioBench against 61.7% before. TestingCatalog also cited the independent Artificial Analysis Intelligence Index, where Fable 5.2 scored 65 against 60 for Gemini 4 Carbon.
What changes for enterprise agent builds?
The most practical change is in how Anthropic wants customers to compose models. Fable 5.2 can act as an advisor to smaller Claude models or as an orchestrator that hands subtasks to Haiku, Sonnet or Opus, Let's Data Science and TestingCatalog reported, and SD Times said it "orchestrates agent teams." Anthropic's own guidance, quoted by 9to5Mac, is to start with Opus 5.5 as the everyday model and move up to Fable or Mythos only when a task demands it, or down to Sonnet 5.5 or Haiku 5.5 when it does not.
That framing invites a tiered design where the expensive model plans and checks while cheaper models do most of the token-heavy work. Anthropic said that at a high effort setting Fable 5.2 roughly matches Opus 5.5 at maximum effort on FrontierCode 2.0 at about half the cost per task, according to a company post quoted by TestingCatalog. If that holds on real workloads, the per-outcome cost of complex coding agents could fall even for teams that move some work onto the more expensive model. TestingCatalog also reported that Anthropic made medium effort the default in Claude apps and Claude Code, a setting that teams will want to review in their own deployments.
Early users named by TestingCatalog include Vercel, Shopify and Datadog, which described the model as strong on ambiguous engineering work.
What should data and AI leaders do now?
Three steps follow from the release. First, rerun internal evaluations rather than relying on vendor scores, paying attention to cost per completed task and not only accuracy, since effort settings now change both. Second, review routing rules in agent platforms so that Fable is reserved for planning, review and hard problems while smaller models handle volume work. Third, check which cloud each team uses to reach Anthropic models, because Fable's availability on AWS, Google Cloud and Azure means procurement, data residency and logging can follow existing cloud agreements rather than a separate contract.
Security teams should also take note of Mythos. Its gains on vulnerability rediscovery are the kind of capability that helps defenders find weaknesses faster, and Anthropic's decision to keep it behind a verification program is a reminder that the most capable cyber models are being gated deliberately. Organizations that want access will need to apply and show a defensive use case.
Frequently asked questions
What is Claude Fable 5.2?
It is Anthropic's frontier model for long, open-ended technical work, released November 9, 2026, with output up to 30% faster than Fable 5.1.
How much does Claude Fable 5.2 cost?
Standard pricing is $8 per million input tokens and $40 per million output tokens, and an optional fast mode costs $16 and $80.
Who can use Claude Mythos 5.2?
Only organizations approved through Anthropic's Life Sciences and Cyber Verification Programs.
Disclosure: AI Enterprise Magazine uses AI tools in its editorial process, including Anthropic's Claude models. Anthropic had no role in this article.
Sources
Filed under: Agentic AI, AI Strategy & ROI, Security & Fraud
