Claude Fable 5.2 Released: Full Breakdown of New Features, Benchmarks and Pricing

Anthropic released Claude Fable 5.2 on November 9, 2026. The update is aimed at difficult, long-running work: codebase-scale engineering, tool-using agents, scientific analysis, visual interpretation, and workflows where a plausible answer is not enough—the result has to survive execution or verification.

The short version is compelling. Fable 5.2 supports a one-million-token context window and up to 128,000 output tokens, produces output up to 30% faster than Fable 5.1, and costs $8 per million input tokens and $40 per million output tokens through the Claude API. An optional fast mode doubles those token rates in exchange for output generation of up to 2.5 times the standard speed.

Claude Fable 5.2 specifications at a glance

| Item | Claude Fable 5.2 | |---|---| | Model ID | claude-fable-5-2 | | Context window | 1M tokens | | Maximum output | 128K tokens | | Standard API price | $8 input / $40 output per 1M tokens | | Fast mode price | $16 input / $80 output per 1M tokens | | Inputs and outputs | Text and images in; text out | | Knowledge cutoff | June 2026 | | Release date | November 9, 2026 |

Adaptive thinking is always enabled, and the default effort level is high. That matters when comparing cost or latency with another model: identical prompts are not necessarily identical experiments if the effort settings differ.

What actually changed

Faster output without switching to a smaller model

Anthropic reports up to 30% faster output than Fable 5.1. Fast mode goes further, increasing output tokens per second while preserving the same underlying model intelligence. It is useful for interactive coding loops, live agents, and time-sensitive analysis. It does not promise an equivalent improvement in time to first token, so teams should measure the latency users actually experience.

A substantial price reduction

Fable 5.1 launched at $10 per million input tokens and $50 per million output tokens. Fable 5.2 reduces both figures by 20%. Prompt caching can lower repeat-input costs further: a cache read is listed at $0.20 per million tokens, while five-minute and one-hour cache writes cost $10 and $16 respectively. Batch processing offers a 50% input-and-output discount for eligible asynchronous jobs.

Better performance on executable work

The most useful release results are not trivia-style tests. On FrontierCode 2.0, which evaluates complex repository-level tasks, Fable 5.2 scored 49.1% at xhigh effort, compared with 32.4% for Fable 5.1. It reached 69.9% on FrontierSWE v2, 64.9% on RoboDojo Sim, and 74.6% on Terminal-Bench-Science 0.1 in Anthropic's published evaluations.

Those numbers suggest meaningful progress, but they are not a warranty for a private codebase. Tool configuration, prompt scaffolding, effort level, pass@k rules, and test infrastructure can move results considerably. Treat public benchmarks as a reason to run an internal trial, not as a substitute for one.

More usable safety controls

Fable 5.2 carries stronger safeguards for advanced biology and cybersecurity work. Anthropic also reports approximately 55% fewer blocks than Fable 5.1 in early internal testing across Claude products, including reductions in both cyber- and biology-related false positives. Some blocked requests can be routed to safer fallback models rather than ending the conversation outright.

Availability

Fable 5.2 is available through the Claude API and Claude products, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Provider-specific model names, regions, quotas, billing arrangements, and preview status differ. In Microsoft Foundry, fast mode is a separate deployment named claude-fable-5-2-fast; on the Claude API it is requested with speed: "fast".

Who should use Fable 5.2?

It is a strong candidate when failures are expensive and the job requires long-horizon reasoning: a multi-package migration, a research workflow with tools, an agent that must recover from execution errors, or analysis spanning large documents and images. It is less attractive for short classifications, routine rewriting, or high-volume extraction that a smaller model can already perform reliably.

Anthropic itself recommends starting most workloads with Claude Opus 5.5, then moving to Fable when evaluations show that the less expensive model falls short. That is sound operational advice. Model choice should be based on cost per successful task, not prestige or token price alone.

For creative teams, one practical workflow is to use Fable for research, technical planning, or script diagnostics, then take approved material into Elser AI for browser-based creation workflows or organize a longer production in ElserStudio. Confirm current provider and model availability before designing an automated handoff.

Final assessment

Claude Fable 5.2 is not simply a faster Fable 5.1. It combines a lower standard price, quicker generation, stronger performance on demanding executable tasks, and revised safeguards. The strongest reason to adopt it is a measured improvement on your own hardest work. Build a representative evaluation set, fix the effort level, record end-to-end latency and review time, and compare total successful-task cost with Opus 5.5 and your current production model.

Frequently asked questions

Is Claude Fable 5.2 generally available?

It is generally available through Anthropic's platform. Availability and lifecycle labels on cloud marketplaces can differ; Microsoft Foundry currently documents Anthropic-hosted Fable deployments as preview.