Claude Fable 5.2 pushes Anthropic toward heavy technical tasks

Anthropic launches Claude Fable 5.2 and Claude Mythos 5.2, two models designed for coding, research, and long-term work. The public version arrives with enhanced safeguards, while Mythos remains reserved for verified organizations.

Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2, two new AI models designed for the most demanding technical tasks, on November 9, 2026. The company says they are faster than the 5.1 versions while improving code, reasoning, and vision. Fable 5.2 is available now to the general public, while Mythos 5.2 remains reserved for verified organizations. This launch is significant as it puts Anthropic back in the race for the best model for heavy professional use, with a very clear focus: work fast, but not just any way.

Models tailored for code, science, and long-term projects

Anthropic presents Fable 5.2 as a model capable of tackling difficult software engineering problems, orchestrating teams of agents, and managing long-term projects. Mythos 5.2 shares the same base but without the same access restrictions. In internal tests and pre-publications cited by the company, the models helped carry out software migrations considered otherwise impossible, design three functional electronic devices, and make progress on robotics evaluations.

The example that draws the most attention remains mathematical. Anthropic claims that about 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in Lean within 30 hours, a language that allows a computer to verify each step. The company specifies that this version used one-third of the output tokens of the first internal trial and produced a proof seven times shorter. For fans of measured promises, this is the kind of demonstration that raises an eyebrow, then two.

What benchmarks say, and what they don’t

On the Artificial Analysis Intelligence Index v5 ranking, Fable 5.2 takes the lead with 65 points at the highest computation level, five points ahead of Gemini 4 Carbon and six ahead of Claude Opus 5.5, according to data cited by The Decoder. Anthropic and the media relaying the announcement also mention a faster output, up to 30% faster than Fable 5.1 and Mythos 5.1, with a decrease in the price displayed per million tokens.

However, the figures should be read with caution. On some sub-tests, Fable 5.2 clearly dominates; on others, it is behind rivals like Kimi K3.1 Pro, Gemini 4 Carbon, or Sonnet 5.5. In other words, the hierarchy varies depending on the type of task, the level of computation required, and the benchmark used. This is a detail that is not trivial for technical teams, as a model can shine in theory and be very costly in production.

Security at the heart of the split between Fable and Mythos

Anthropic emphasizes one point: Fable 5.2 blocks sensitive tasks in cybersecurity and biology, while Mythos 5.2 removes some of these restrictions for organizations that have gone through its verification programs. According to the company, this separation aims to limit risky uses, notably penetration testing and exploit code generation. Mythos 5.2 would also be its most capable model for cybersecurity, with results superior to previous versions in vulnerability discovery.

On its internal CVE Rediscovery test, Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1 in an October 2026 campaign covering 1,829 CVEs. At the same time, the company claims to have deliberately slowed down certain capabilities that could directly accelerate AI development, linking this to the debate on the risk of system self-improvement. The boundary between performance and safeguards remains the place where Anthropic wants to be judged.

Why this launch matters in the model battle

The timing is no coincidence. According to The Decoder, these models arrive shortly before Anthropic’s IPO, in a context where every performance gain also serves as a strategic signal. For companies, the message is clear: Anthropic wants to remain a leading provider for software development, applied research, and multi-agent workflows. For developers, this mainly means one very prosaic thing: reassessing the real cost of a task, not just what the model promises on paper.

The real novelty may not be the raw score, but the way Anthropic links performance, security, and orchestration of multiple agents. This is where part of the 2026 market is played: less on the "chatbot that knows everything," more on the system that manages a project without getting lost along the way.

Key points

  • Fable 5.2 has been available since November 9, 2026.
  • Mythos 5.2 remains reserved for verified organizations.
  • Anthropic announces up to 30% additional speed.
  • Fable 5.2 scores 65 points on Artificial Analysis v5.
  • 1,000 agents formalized Fermat in 30 hours.
  • Mythos 5.2 found 10.2% more CVEs.

By the numbers

  • 65 points — Fable 5.2 score on the Artificial Analysis Intelligence Index v5, according to The Decoder.
  • 30% — speed gain announced compared to Fable 5.1 and Mythos 5.1, according to Anthropic.
  • 1,829 CVEs — base of the internal CVE Rediscovery test, according to Anthropic.
  • 30 hours — announced duration to formalize the proof of Fermat's Last Theorem.
  • 10.2% — increase in vulnerabilities found by Mythos 5.2 compared to Opus 5.5, according to Anthropic.

To read