Claude Fable 5.2
Next generation of intelligence for the hardest knowledge work and coding problems.
Announcements
Introducing Claude Fable 5.2
Our most capable model for technical work, setting a new bar in coding, finance, and the sciences.
View moreIntroducing Claude Fable 5.1
Our most capable model for coding and knowledge work, with research capabilities that offer an early glimpse of how AI will soon contribute to scientific progress.
View moreImproving Fable 5’s biology safeguards
We’re making updates to Claude Fable 5’s biology safeguards in a way that substantially reduces false positives.
View moreClaude Fable 5 is rolling out
Access to Claude Fable 5 has been restored. It brings 5th-generation intelligence to your most ambitious coding and professional work.
View moreClaude Fable 5 access unavailable
We apologize for this disruption to our customers and are working to restore access as soon as possible.
View moreClaude Fable 5
Claude Fable 5 introduces our 5th model generation for your most ambitious work. Tackle days-long, complex, and asynchronous tasks previous models couldn’t sustain.
View more
Availability and pricing
For individuals and organizations taking on their hardest technical work, Claude Fable 5.2 is available to Pro, Max, Ultra, Team, and Enterprise users.
For developers building solutions that demand our highest level of intelligence, Claude Fable 5.2 is available on the Claude Platform natively, through available marketplaces, and in Amazon Web Services, Google Cloud, and Microsoft Foundry.
Claude Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. Cache reads cost $0.20 per million tokens. To learn more, check out our pricing page. To get started, use claude-fable-5-2 via the Claude API.
Fast mode for Claude Fable 5.2 is available on the Claude Platform, Claude Models on Google Cloud (Anthropic managed and Google managed), and Microsoft Foundry (Anthropic hosted). It’s also available to use on Claude Code. It delivers up to 2.5x faster speed, and costs $16 per million input tokens and $80 per million output tokens.
For workloads that need to run in the US, US-only inference is available at 1.1x pricing for input and output tokens. Learn more.
Use cases
Claude Fable 5.2 is our most capable model built for your hardest technical work. It takes on ambiguous problems for days, questions and verifies its own work, and orchestrates subagents to improve results in a loop.
Coding
Claude Fable 5.2 is our most capable model for ambitious software engineering. It takes on large, loosely defined problems across multiple repositories and large monorepos, brings strong architectural judgment, and tests its own work as it goes. It’s also more willing to question inherited conclusions, so it finds root causes our previous models miss.
Agents
Claude Fable 5.2 is our strongest lead agent. In our testing, we found it can run ambiguous engineering work for days, coordinate hundreds or even more than a thousand subagents, and use tools more efficiently along the way. It keeps itself honest on long runs by building controls and flagging results that don’t reproduce.
Scientific research
Claude Fable 5.2 can come up with new analytical methods for difficult physics computations, and engineer fully functional electronic devices. It also improves the efficiency of tasks that researchers already do, such as formalizing mathematical proofs.
Data analysis and machine learning
Claude Fable 5.2 is well suited to research loops where the goal is to measure, try something, and improve. It runs ablations, hill-climbs machine learning metrics, and verifies that each change did what it expected, without much pushing.
Financial analysis
Claude Fable 5.2 brings deep reasoning to quantitative finance, from multi-agent research to forecasting and analysis. It catches subtle, second-order edge cases, and builds reports that could take a week of effort.
Design for 3D, hardware, and motion
Claude Fable 5.2 is our best model for designing objects for the real world and on screens. It reads detailed images and PDFs, builds interactive 3D worlds, designs hardware components, creates simulations, animations, and educational videos with sound in code, and can manipulate environments in robotics.
Safeguards
Claude Fable 5.2 includes robust safeguards for cybersecurity and biology. Many queries in these domains are automatically routed to less capable models if flagged by these safeguards. Learn more about how the fallback experience works. Teams whose legitimate work is affected can apply to our verification programs for cybersecurity and life sciences research for adjusted safeguards.
We extensively test and evaluate our models to ensure they meet Anthropic’s standards for safety, security, and reliability. The accompanying system card covers safety results in depth.
Data retention
Using Fable models requires 30-day data retention for safety monitoring by default. With Enterprise Frontier Safeguards (EFS), automated safety monitoring is run on prompts and responses that are stored in a customer’s infrastructure, making it compatible with zero data retention (ZDR). Until EFS is available, eligible customers can continue using Claude Fable 5.2 with ZDR before implementing safety monitoring with EFS.
Learn more about data retention. Learn more about Enterprise Frontier Safeguards.
Benchmarks
Claude Fable 5.2 sets a new bar in coding and the sciences.
| Fable 5.2 | Fable 5.1 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon | |
|---|---|---|---|---|---|
| Agentic codingFrontierCode 2.0 | 49.1%Xhigh effort | 32.4%Xhigh effort | 45.5% | 42.8% | 41.1%Medium effort |
| Agentic codingFrontierSWE v2 | 69.9% | 56.3% | 62.3% | 65.5% | 61.2% |
| Business workflowsAutomationBench | 51.4% | 31.4% | 42.5% | 41.4% | 53.9% |
| Agentic scientific researchTerminal-Bench-Science 0.1 | 74.6%75.3% (Mythos 5.2) | 52.6% | 58.7% | 68.1% | 64.8% |
| CybersecurityCyScenarioBench† | 76.3%Mythos 5.2 | 61.7%Mythos 5.1 | 67.6%without safeguards | 61%GPT-6 Cyber | — |
| CybersecurityCVE Rediscovery | 89.6%Mythos 5.2 | 78.3%Mythos 5.1 | 81.3%without safeguards | — | — |
| Protein designDe novo binder design (24 hr) | 85.1%Mythos 5.2 | 79.1%Mythos 5.1 | 82.6%without safeguards | — | — |
| Agentic searchWANDR (relaxed score) | 82.1% | 68.7% | 72.3% | — | — |
| Computer useOSWorld 2.1 (offline set, partial) | 89.3% | 83.5% | 87.2% | 79.5% | 81.5% |
| RoboticsRoboDojo Sim‡ | 64.9% | 27.6% | 39.4% | 29.8% | — |
| Knowledge workAA-Briefcase v1.1 | 66.1%rubric | 61.5%rubric | 61.2%rubric | 52.0%rubric | 61.8%rubric |
| 1785Elo | 1669Elo | 1800Elo | 1569Elo | 1611Elo | |
| Knowledge workRemote Labor Index (RLI) | 41.3%High effort | 17.9% | 21.3% | 20.8% | — |
| Multidisciplinary reasoningHumanity’s Last Exam (with tools) | 74.2% | 65.6% | 67.7% | 57.2% | — |
| Visual chart recognitionChartography (no tools) | 73.4% | 44.8% | 64.4% | 71.0% | 72.5% |
For details on evaluation methodology, see the Claude Fable 5.2 and Claude Mythos 5.2 System Card. Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. † On CyScenarioBench, each lab’s scores are based on unique tasks. ‡ RoboDojo Sim shows the highest-available effort and was run internally.
Hear from our customers
“The model is in my head. It understands the intent behind every prompt I’ve given it. I haven’t had to say ‘No, I meant…’ a single time. Projects we’ve been pushing along with other models for weeks got finished and approved by our teammates in a few days with Claude Fable 5.2, and the humans reviewing its code had less to say.”
“You can just throw Claude Fable 5.2 at new, ambiguous systems it has never seen before, and it will figure out how to actually work within them. It’s almost like an incident commander that can send out faster models to get quick signals and coordinate, which removes that burden from me as an operator.”
“Claude Fable 5.2 says it doesn’t know things instead of hallucinating. At one point it told me its last estimate was wrong and that it wouldn’t give a number until it had measured the change. Very impressive!”
“In an I-9 compliance review, Claude Fable 5.2 gave the client clearer direction than Fable 5.1. It identified the employees needing immediate attention and explained how to correct each issue and who should act. It’s an exciting improvement in the model’s ability to give practical legal guidance.”
“Claude Fable 5.2 can complete tasks that previous models struggled with. The clearest examples so far are building higher quality 3D scenes from scratch and rebuilding an experience from a visual reference. It’s very strong on the harder game-dev tasks, where it keeps iterating and improving instead of getting stuck.”
“Claude Fable 5.2 is great for taking a loose idea and refining it. Its answers are really well considered, and within a few turns I can almost always say ‘ok, go do that,’ and trust that it’ll complete it with little intervention. It feels like working with another staff-level engineer.”
“Claude Fable 5.2 is the strongest model we have measured on data-eng-bench, our open-source benchmark of 103 data engineering tasks. It solved tasks consistently across repeated runs and passed every interactive coding task we gave it.”
“In a single session, Claude Fable 5.2 built a hypervisor, the software that walls off virtual computers sharing one machine. Its core is mathematically proven to keep them apart. More than 3,500 machine-checked proofs cover every case, not just the ones you test. It’s the first model I’ve seen break a problem this large into pieces it can prove, fit them together, and keep taking on more ambitious work. A very clear upgrade from Fable 5.1.”
“In a recent evaluation, we asked Claude Fable 5.2 to investigate how one organization tested a new integration in its core technology systems. It traced the answer to the right sources and didn’t overstate what the evidence showed. It also flagged that the question’s framing could be misleading. In investment research, that restraint makes clear what remains unknown and where the market may be mispricing risk or opportunity.”
“Before putting AI to work across workflows, you need to trust its approach and conclusions. In our testing, Claude Fable 5.2 identified inconsistencies, drew on context across entire document sets, and adjusted recommendations based on relevant evidence. For example, when reviewing files attached to a loan application, it flagged that the salary deposits and the bank statement’s balance didn’t match, so the underwriter could focus on what to check.”
“I had Claude Fable 5.2 run a 10–12 hour security audit. It sent out about 170 reviewer agents and came back with a write-up, fix pull requests and a policy proposal. It works until the job is actually complete instead of stopping at a plausible checkpoint. Its self-correction and honesty about gaps impressed me. At high effort, I’d use this over what I run today for long unattended audits and doc-heavy work.”
“With Fable 5.1 we saw the first glimmers of a model coming up with truly innovative research ideas. Claude Fable 5.2 continues that trend. It had the same insight we were impressed with from Fable 5.1 and then came up with a second, entirely new angle. It achieved new best scores on three of the four quant research problems we’ve finished. We’d like to use it for high-impact, long-running research tasks, or for coordinating teams of other models.”
“In our quantitative research work, Claude Fable 5.2 delivered very strong improvements vs Fable 5.1, especially in the setup with multiple coordinating agents. Its maths research capabilities represent another meaningful advance too.”
“Phonon calculations are one of the most expensive routine steps in computational materials science, which is why labs often skip stability checks when screening new materials. Claude Fable 5.2 made Quantum ESPRESSO’s phonon code about 36x faster on average than stock on the same GPU, taking our 13 benchmark cases from 3.3 hours to under 4 minutes. That beats Fable 5.1’s speedup by 3–4x.”
Frequently asked questions
We offer Claude models across the spectrum of speed, price, and performance. Claude Fable 5.2 is our most capable generally available model for technical fields. It’s best for ambitious, open-ended technical work in coding, scientific research, and finance, where judgment and accuracy matter most. For most everyday coding and knowledge work, we recommend starting with Claude Opus 5.5.
Making a model as capable as Fable 5.2 generally available comes with risks. The model’s capabilities in specific areas like cybersecurity, biology, and chemistry are advanced enough that they could be misused to create wide-reaching cyberattacks or build dangerous bioweapons. For that reason, Fable 5.2 comes with safeguards that block or limit its performance in these areas. Many queries on topics including biology and cybersecurity will instead receive a response from Opus models.
Security teams can apply for advanced cyber capabilities on Claude Mythos 5.2, Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and future models through the Cyber Verification Program. Life sciences researchers can apply for advanced capabilities through the Life Sciences Verification Program.
For most Claude applications, queries flagged by our cybersecurity safeguards automatically route to Opus 4.8 and biology safeguards route to Opus 5. API customers must configure their settings with our new Fallback API. Learn more
Pricing depends on how you want to use Claude Fable 5.2. To learn more, check out our pricing page.