Anthropic says Fable 5.2 nearly doubles its predecessor’s hardware-design score. Token prices fell 20%, but the cost per task doubled at full effort.

Anthropic’s newest flagship AI model, Fable 5.2, scored 83.5 on a hardware-design benchmark, nearly double its predecessor’s 44.4, according to the company. The model, which Anthropic made generally available Nov. 9, is priced 20% lower per token than Fable 5.1, at $8 per million input tokens and $40 per million output tokens.

At Max effort, Fable 5.2 used about 1.7 times as many output tokens as its predecessor in Artificial Analysis’s testing, a prominent independent tester of AI models. The firm measured $26.54 per Intelligence Index task for Fable 5.2 at Max, compared with $12.99 for Fable 5.1 at the same setting. Fable 5.2 scored 62 at both High and Xhigh, at $9.08 and $14.83 per task, respectively. Both settings scored ahead of every other model on the index.

At the Medium effort setting, which is the default in Claude’s apps and Claude Code, Fable 5.2 completes difficult tasks with fewer tokens than its predecessor Fable 5.1, according to Anthropic. On the FrontierCode 2.0 benchmark, which measures whether maintainers of open-source projects would merge a model’s code changes, Fable 5.2 at High scored 45.3%, just under Opus 5.5’s best of 45.5%, for roughly half the cost per task.

Anthropic also tested the model on physical products. Working from short descriptions, Fable 5.2 designed the electronics, enclosures and firmware for a voice-assistant hub, a USB speakerphone and a handheld virtual pet. Each circuit board was ready for manufacturing in under a day, according to Anthropic. Diode, a company that designs and manufactures circuit boards, built the prototypes, which worked without electrical modifications.

The devices remain early prototypes, with electrical validation, antenna tuning, thermal testing and certification still pending. Diode’s engineers handled the equipment, assembled the boards and ran the firmware under the model’s direction. The work follows OpenAI’s recent account of GPT-6 Astra designing a talking Codex pet, which took several manufacturing review and iteration rounds.

BenchCAD 2.1, a public benchmark, asks AI models to rebuild 3D parts, assemblies and circuit boards from engineering drawings and rendered images. Opus 5.5 scored 65.3 on the same test. On PCB Gym, an internal circuit-board benchmark that Diode ran, Fable 5.2 hit 69%, compared with 25% for Fable 5.

On RoboDojo Sim, which tests control of a simulated two-arm robot from camera images, Fable 5.2 scored 64.9%, more than double Fable 5.1’s 27.6% and GPT-6 Astra’s 29.8%, according to Anthropic’s internal runs.

Sources: Anthropic launch and research posts, Artificial Analysis; read Nov. 9, 2026. Scores are at each model’s highest available effort setting unless noted.

At Max, Fable 5.2 led Artificial Analysis’s Intelligence Index, scoring 65 to 60 for Google’s Gemini 4 Carbon. Google announced Carbon on Oct. 28, initially making it available to selected cybersecurity defenders.

Anthropic still recommends Opus 5.5 for most coding and knowledge work and suggests using Fable for harder, open-ended projects, including ones where it can delegate simpler tasks to smaller models. Fable 5.2’s safeguards block most cybersecurity tasks and much of research biology. Mythos 5.2 uses the same underlying model and is available through Anthropic’s verification programs for cybersecurity and life sciences.