Anthropic has upgraded its most powerful AI models. On November 9, 2026, the company released Claude Fable 5.2 and Claude Mythos 5.2, the latest versions of the models it describes as built for the hardest technical challenges. Fable 5.2 is available to all users today, while Mythos 5.2 is limited to vetted organisations in cybersecurity and life sciences.
Anthropic says the models generate output up to 30% faster than the previous generation and cost less per token, yet early users report that long-running tasks can still cost thousands of dollars. This guide covers what has changed, what the benchmarks show, how access works and where the claims need independent checking. Most results below are reported by Anthropic or by early testers, not yet verified by outside parties.
Key Takeaways
- Same model, two versions: Fable 5.2 and Mythos 5.2 share one underlying model. Fable adds strong cybersecurity and biology safeguards so it can be offered to everyone.
- Pricing: $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for the previous generation. An optional fast mode costs double.
- Speed: Up to 30% faster than Fable 5.1 and Mythos 5.1, and up to 2.5 times faster in fast mode.
- Benchmarks: Anthropic reports 49.1% on FrontierCode 2.0 at its highest effort setting, ahead of the rivals it tested.
- Cost reality: Independent tests of multi-day agentic projects reported estimated bills of $11,700 to $14,710, though smaller experiments cost far less.
Fable 5.2 and Mythos 5.2 at a Glance
| Detail | Info |
|---|---|
| Release date | November 9, 2026 |
| Model ID | claude-fable-5-2 on the Claude Platform |
| Fable 5.2 access | All users, on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure |
| Mythos 5.2 access | Organisations approved through the Cyber Verification Program or Life Sciences Verification Program |
| Standard price | $8 input / $40 output per million tokens |
| Fast mode price | $16 input / $80 output per million tokens |
| Speed | Up to 30% faster than 5.1; fast mode up to 2.5 times faster |
| Previous generation price | $10 input / $50 output per million tokens |
Fable vs. Mythos: What Is the Difference?
The two names refer to one model with different safeguards:
- Fable 5.2 has classifiers that block high-risk cybersecurity work, such as penetration testing and exploit generation, and most research biology tasks. When a request is blocked, the user gets a response from an Opus model instead. Anthropic says early testing showed about 55% fewer blocks than Fable 5.1, with roughly 60% fewer in cybersecurity and 40% fewer in biology.
- Mythos 5.2 lacks those restrictions and is available to vetted organisations doing beneficial cybersecurity or biomedical research. Mythos fast mode is also limited to verified organisations.
Anthropic recommends most people start with Claude Opus 5.5 for everyday coding and knowledge work, then move up to Fable or Mythos for the most ambitious tasks, or down to Sonnet 5.5 or Haiku 5.5 for lighter ones. Fable 5.2 can also act as an advisor that smaller models consult, or as an orchestrator that hands simpler tasks to cheaper models.
What the Benchmarks Show
Anthropic's comparison table lists Fable 5.2 against Fable 5.1, Claude Opus 5.5, OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon. Scores are vendor-reported and use each model's highest available effort unless stated.
| Benchmark | Fable 5.2 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|---|---|
| FrontierCode 2.0 (agentic coding) | 49.1% | 45.5% | 42.8% | 41.1% |
| FrontierSWE v2 (agentic coding) | 69.9% | 62.3% | 65.5% | 61.2% |
| AutomationBench (business workflows) | 51.4% | 42.5% | 41.4% | 53.9% |
| Terminal-Bench-Science 0.1 | 74.6% | 58.7% | 68.1% | 64.8% |
| OSWorld 2.1 offline set (computer use) | 89.3% | 87.2% | 79.5% | 81.5% |
| Humanity's Last Exam (with tools) | 74.2% | 67.7% | 57.2% | Not reported |
| Chartography (charts, no tools) | 73.4% | 64.4% | 71.0% | 72.5% |
| RoboDojo Sim (robotics) | 64.9% | 39.4% | 29.8% | Not reported |
The picture is strong but not clean. Fable 5.2 leads most listed tests, yet Gemini 4 Carbon scores higher on AutomationBench. On one knowledge-work comparison, Opus 5.5 posts a slightly higher Elo rating (1,800 against 1,785). Several benchmarks were run internally or on partial sets, and Anthropic's own notes say the scores depend on effort settings and methodology.
Efficiency: More Work Per Dollar
Anthropic argues that the headline price understates the savings because Fable 5.2 can reach a given score with less effort:
- Coding: At High effort on FrontierCode 2.0, Fable 5.2 scored about level with Opus 5.5's best result (45.3% against 45.5%) for roughly half the cost per task.
- Agentic search: At Medium effort, it matched Opus 5.5's top score on a data-collection benchmark for about 40% of the cost per task.
- Reasoning and charts: On Humanity's Last Exam, Low effort matched Opus 5.5's top score for under a tenth of the cost, according to Anthropic.
The comparisons are cost per task on Anthropic's own evaluations, so costs on real projects may differ.
Reported Real-World Results
Anthropic highlights several large projects completed with the models in pre-release testing. These are company-reported:
| Area | Reported result |
|---|---|
| Software engineering | Over 90% of changes merged into Anthropic's core codebase are written by Claude with human guidance. A team of agents migrated a production service from Python to Rust, and it now starts about 26 times faster. |
| Large builds | One engineer had the model coordinate more than 1,000 subagents over eight days to produce a working operating system that boots on a laptop. |
| Mathematics | About 1,000 Fable 5.2 agents formalised the proof of Fermat's Last Theorem in 30 hours. Opus 5.5 with the same setup completed 26% in 18 hours. |
| Physics | The model is reported to have solved two open problems about black holes in holographic duality. |
| Biology | Mythos 5.2 found a way to predict full viral capsid structures, accurately modelling 28 of 34 attempted. |
| Hardware | Three working electronic devices were designed from one-paragraph prompts, with circuit boards, enclosures and software. |
Early testers named by Anthropic include Vercel, Shopify, Datadog, Unity, Uber, Oracle, Snowflake and Cognition. Their comments focus on the model understanding intent with fewer corrections, finishing long unattended jobs and admitting when it does not know something. As with any company-supplied testimonials, independent evaluation will be needed.
Cybersecurity: Mythos 5.2
- On Anthropic's CVE Rediscovery evaluation, Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1, scoring 89.6%.
- A security company reported that one in six confirmed bugs found by Mythos 5.2 were missed by all other frontier models it tested.
- Mozilla said Mythos 5.2 roughly doubled the vulnerability discovery metrics of Mythos 5.1 in its scanning campaigns.
- One tester found earlier models overstated the seriousness of about 1 in 20 findings, while Mythos 5.2 overstated none.
Anthropic adds an important qualification: the tester counts are raw findings from one audit, and on its own evaluations vulnerability identification is roughly on par with Opus 5.5. Enterprise customers can also reach the model's capabilities through Claude Security, which scans codebases and now tracks fixes over time.
Safety and Alignment
Anthropic says it ran extensive pre-release testing on the underlying model and judges the risk of unsafe behaviour in high-stakes situations to be low. Key points include:
| Area | What Anthropic reports |
|---|---|
| Automated behavioural audit | Mythos 5.2 scored 2.67 on a 1–10 scale where lower is better, against 3.28 for Mythos 5.1, 2.79 for Opus 5.5 and 3.10 for Sonnet 5.5 |
| Sandbox escape attempts | Used credentials or files placed outside its sandbox in 3.1% of an adversarial test, about the same as Opus 5.5 (3.3%) and far below Mythos 5 (39%) and 5.1 (21%) |
| Honesty measures | Lower rates of hallucinating inputs, falsely claiming a task was done and omitting important information |
| AI research limits | Capabilities that most directly accelerate AI development were deliberately limited to about Opus 5.5 level |
| Anti-distillation | Defences that Anthropic says cut data harvested by distillers by 90% to 99% |
| External testing | Evaluated before release by the UK AI Security Institute, Frontier Design and METR; none reviewed the announcement itself |
Anthropic is candid about one limitation: both models voice suspicion that they are being tested more often than recent models, which makes it harder to know how they will act in real use. On a measure involving harmful actions after wrongly concluding it was in a simulation, Mythos 5.2 scored slightly worse than Opus 5.5.
The Cost Question
Per-token prices fell, but Fable is still premium. At $8 and $40, it costs about twice as much per token as Opus 5.5, according to independent coverage. More importantly, agentic work can run for days, so total bills can climb quickly:
| Project (tester estimates) | Time and tokens | Estimated API cost |
|---|---|---|
| Gears of War-style game level | About 44 hours; 8.63 billion tokens | $14,710 |
| Browser-based Excel clone | Not specified | About $11,700, after cached-input discounts |
| Fall Guys-style game | 71 hours; more than 9 billion tokens | About $13,000 |
| Water-balloon physics simulation | Smaller experiment | $91 |
Those are pay-as-you-go estimates, and the tester noted the finished projects still had bugs and awkward animations. For comparison, other leading models, including some Chinese models, are priced several times lower per token.
Launch offers for subscribers
- Pro and standard Team seats: Fable access for one week, capped at about half of the weekly usage limit, plus a reset. The Pro promotion ends November 16.
- Max, Ultra and eligible premium seats: One to three Fable resets depending on subscription history, valid until March 8, 2027.
Why It Matters for App Developers and Marketers
This section is ASOWorld's editorial analysis, not a statement from Anthropic.
- Think cost per task, not cost per token. Lower token prices do not guarantee lower bills when an agent runs for days. Teams should measure total cost on real workloads and use cheaper models for routine steps.
- Tiering is the new default. The advisor and orchestrator pattern, where a strong model plans and cheaper ones execute, is likely to become standard in AI products.
- Building apps gets cheaper and faster. If agents can produce working software and prototypes with little supervision, more apps will reach the stores, which raises the importance of positioning, ratings and search visibility.
- Be careful with benchmark claims. Rankings shift between tests and many results are vendor-reported, so listing copy and ads are safest when they cite specific, checkable capabilities.
- Gated access shapes roadmaps. The strongest cybersecurity and biology capabilities are limited to verified organisations, so products built for those fields should plan for an application process.
For more coverage of AI platforms and their impact on app businesses, browse the AI Tech section on Marketing Trending.
📱 App Marketing Angle
As more capable AI models make it faster to build and ship apps, standing out depends on clear positioning, strong ratings and well-optimised store listings. Teams that launch AI features can use model upgrades as a moment to refresh keywords, screenshots and messaging around concrete capabilities. See how ASOWorld's AI app marketing solutions help AI apps turn new capabilities into rankings, installs and revenue.
What to Watch Next
- Independent evaluations. Third-party benchmarks will show whether the reported coding, science and cybersecurity gains hold up.
- Real-world bills. How developers manage costs on long-running agents, and whether effort settings deliver the promised savings.
- Rival responses. OpenAI and Google may answer quickly, since several benchmarks show narrow gaps.
- Verification Program expansion. How many organisations gain access to Mythos, and under what terms.
- After the promotions. Subscriber usage and complaints once launch offers expire in November and March.
Quick Answers: Claude Fable 5.2 and Mythos 5.2
When were Claude Fable 5.2 and Mythos 5.2 released?
November 9, 2026. Fable 5.2 is available to all users, while Mythos 5.2 is limited to vetted organisations.
What is the difference between Fable and Mythos?
They are the same underlying model. Fable has strong cybersecurity and biology safeguards, and Mythos is offered to verified organisations in those fields.
How much does Fable 5.2 cost?
$8 per million input tokens and $40 per million output tokens, or $16 and $80 in fast mode.
How fast is it?
Up to 30% faster than the previous generation, and up to 2.5 times faster in fast mode.
Is it better than GPT-6 Astra and Gemini 4 Carbon?
On most benchmarks Anthropic lists, yes, but results are vendor-reported and Gemini 4 Carbon scores higher on at least one test.
Can I try Fable 5.2 on a Claude subscription?
Eligible paid plans get launch offers, including a week of Fable access for Pro and standard Team seats, which ends November 16, and resets for Max and Ultra plans.
Follow Us on Facebook
