AboutAI ToolsCursorClaude CodeLovableBolt.newAI tool reviewsTutorialsVibe codingAI agentsMobile apps with AIAll tutorialsBlogSponsorWork With MeJoin 285K+ Subscribers

14 November 2026 · 3 min read

Anthropic releases Claude Fable 5.2 and Mythos 5.2 with lower pricing and 30% faster output

Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026, cutting API prices and delivering up to 30% faster token generation while targeting ambitious agentic and scientific workloads.

Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026 [4]. The two configurations share a single underlying model but differ in safeguards and access conditions [2][4]. Fable 5.2 is available immediately on all paid Claude plans and the API, while Mythos 5.2 remains restricted to organisations approved through Anthropic’s Life Sciences and Cyber Verification Programmes [1][4].

What is new

The 5.2 models bring several improvements over their 5.1 predecessors. Anthropic reports output generation up to 30% faster [1][3][4]. Benchmark figures show gains across agentic coding, scientific research and robotics simulations: FrontierCode 2.0 rose from 32.4% to 49.1% at Xhigh effort, Terminal-Bench-Science 0.1 from 52.6% to 74.6%, and RoboDojo Sim from 27.6% to 64.9% [4]. On the independent Artificial Analysis Intelligence Index v5.0.1, Fable 5.2 scores 65, compared with 49 for Fable 5.1 and 59 for Opus 5.5 [4].

Performance and benchmarks

Anthropic’s own table places Fable 5.2 ahead of Opus 5.5 on ten of eleven general benchmarks [4]. It does not lead in every category: Gemini 4 Carbon scores 53.9% on AutomationBench against Fable 5.2’s 51.4%, and Opus 5.5 retains a higher Elo on AA-Briefcase v1.1 (1800 vs 1785), although Fable 5.2 wins the rubic half of that test [2][4]. Compared with other models, FrontierCode 2.0 results are 49.1% (Fable 5.2, Xhigh), 45.5% (Opus 5.5), 42.8% (GPT-6 Astra) and 41.1% (Gemini 4 Carbon, Medium) [4]. Where effort settings differ—Medium is the default in the Claude apps and Max/High are API settings—cross-model comparisons require care [2][4].

Anthropic also showcased two ambitious demonstrations: 1,000 Fable agents formalised Fermat’s Last Theorem in 30 hours of machine time, and the model produced a bootable operating system after coordinating over 1,000 agents for eight days [3][4]. The Mythos configuration, with fewer safeguards, correctly modelled 28 of 34 natural and engineered capsids in a biology evaluation and scored 89.6% on a CVE Rediscovery benchmark for vulnerability detection [2][3].

Pricing and fast mode

API prices have been cut from $10/$50 to $8 per million input tokens and $40 per million output tokens, with cached reads now $0.20 per million tokens [4]. A new optional fast mode, offering up to 2.5× faster token generation, costs $16 per million input tokens and $80 per million output tokens [4]. Fast mode is available on Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry, but cannot be combined with the Batch API [4]. For comparison, standard Opus 5.5 is $4/$20, and Opus 5.5 fast mode matches Fable 5.2’s standard rate at $8/$40 [4].

Availability and safeguards

Fable 5.2 runs on Claude channels with safety classifiers that block high-risk cybersecurity tasks (including penetration testing and exploit generation) and most research-biology work [2][4]. When a request is blocked, the system falls back to Opus 4.8 for cybersecurity or Opus 5 for biology [2]. Mythos 5.2 lifts those blocks but is only available to vetted organisations, and Enterprise customers can also access it through Claude Security [3][4]. Anthropic says Fable 5.2 blocks roughly 55% fewer requests than Fable 5.1 in early testing, though it still advises most developers to start with Opus 5.5 for everyday coding and knowledge work [4].

What it means for developers

The price drop and performance lift make Fable 5.2 more practical for complex, long-running agentic tasks, but the official recommendation remains clear: Opus 5.5 is the daily driver [1][4]. For builders experimenting with agents that coordinate multiple models or run for hours, Fable 5.2 offers stronger tool use and reasoning at half the cost of Opus 5.5 when run in fast mode, though the standard Fable 5.2 still costs twice as much as Opus 5.5 [4]. The fast-mode option gives teams a way to trade cost for latency when response time matters. Developers integrating the API should note that the default effort is High for Fable 5.2, while the Claude apps and Claude Code default to Medium, which influences both quality and cost [4].

With the 5.2 pair, Anthropic maintains a deliberate separation: a broadly accessible model with visible guardrails and a specialist configuration that skips those blocks for approved security and life-sciences labs [2][4]. The shared base model means developers testing on Fable 5.2 today are exercising the same capabilities available to verified organisations under Mythos 5.2, albeit with the protective classifiers in place.

Sources

  1. Anthropic introduces Claude Fable 5.2, Mythos 5.2
  2. Claude Fable 5.2 and Mythos 5.2: What’s Different | NeoTeo
  3. Claude Fable 5.2 and Mythos 5.2: Benchmarks and Pricing
  • claude
  • anthropic
  • ai models
  • api pricing
  • developer tools

Researched and drafted with AI from the sources above.

Learn to build with AI

Hands-on tutorials on Cursor, Claude Code, AI agents and more.