Articles

Claude Fable 5.2 and Mythos 5.2: Pricing, Benchmarks, Access

Anthropic released Claude Fable 5.2 and the vetted-only Mythos 5.2, cutting token prices 20% to $8/$40 and claiming 30% faster output. What changed.

Chisato Chisato · · 6 min read
A stylized Claude model badge glowing against a dark background

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9, 2026, the third paired release in the Fable/Mythos line and the second point update since the generation debuted in June. As with the previous two releases, the company describes the pair as one underlying model in two configurations: Fable 5.2 ships to all customers with cybersecurity and biology safeguards switched on, while Mythos 5.2 drops some of those safeguards and is restricted to vetted organizations.

The headline change for most customers is cost. Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens, down 20% from Fable 5.1’s $10 and $50. Mythos 5.2 carries the same rates. Anthropic also says both models generate output up to 30% faster than the 5.1 generation, and it is offering a fast mode that runs up to 2.5x faster at $16 / $80 per million tokens.

What Anthropic shipped

Fable 5.2 is generally available through the Claude Platform API and through Amazon Web Services, Google Cloud and Microsoft Azure, according to Anthropic’s release materials and coverage of the launch. Anthropic’s platform documentation lists a 1 million token context window, which is both the default and the maximum, and up to 128,000 output tokens per response.

Mythos 5.2 is available only to organizations cleared through Anthropic’s Cyber Verification Program and Life Sciences Verification Program, the same trusted-access tracks the company used for Mythos 5.1. Admission requires use-case review and monitoring commitments, which keeps the less restricted model with security researchers and life-sciences groups.

The pricing move continues a pattern. The 5.1 release in September held headline token rates flat and cut cached-input reads by 75%. The 5.2 release cuts the headline rates directly. Coverage of the launch noted that even after the cut, Fable 5.2 still costs about twice as much per token as Claude Opus 5.5, the company’s most recent Opus model.

The benchmarks

Anthropic’s launch tables, as reported by several outlets, put Fable 5.2 ahead of Opus 5.5 on most of the general benchmarks the company published. One analysis of the launch said Fable 5.2 led Opus 5.5 on ten of eleven of Anthropic’s general benchmarks, and reported a score of 65 against 59 on the Artificial Analysis Intelligence Index v5.0.1.

Figures reported from Anthropic’s own tables include:

  • SWE-bench Pro: 93.0% for Fable 5.2, versus 89.9% for Opus 5.5.
  • FrontierCode 2.0: 49.1% for Fable 5.2, versus 45.5% for Opus 5.5.
  • OSWorld 2.1: 89.3% for Fable 5.2, versus 87.2% for Opus 5.5.
  • SWE-bench Multimodal: 74.1% for Fable 5.2, against 54.7% for Fable 5.1 and 61.4% for Opus 5.5.
  • Terminal-Bench-Science 0.1: 74.6% for Fable 5.2 and 75.3% for Mythos 5.2.

The model’s weakest reported result was AutomationBench, at 51.4%, behind Google’s Gemini 4 Carbon at 53.9%, according to one review of the launch numbers.

Independent testing does not line up with every vendor figure. Evaluation firm Vals.ai reported Fable 5.2 at 62.29% on SWE-bench Multimodal, ranking it second of 22 models and behind Opus 5.5 at 70.21%. That reverses the order in Anthropic’s table. Vals.ai also placed Fable 5.2 first of 63 models on Terminal-Bench 4.0 at 65.15%, tied with Opus 5.5. Other coverage reported different Terminal-Bench 4.0 numbers for both Fable and Mythos. The differences likely come from harness settings and evaluation setups that vary between testers, and none of the reports reconciled them.

Mythos 5.2 and the cyber numbers

The capability gap between the two configurations shows up most clearly in security evaluations. Anthropic reported that Mythos 5.2 scored 89.6% on a CVE Rediscovery evaluation, which tests whether a model can independently find known vulnerabilities in real codebases. Mythos 5.1 scored 78.3% on the same test. The company also reported 76.3% for Mythos 5.2 on CyScenarioBench, a scenario-based cyber evaluation. Coverage of the launch did not include a matching Fable 5.2 figure on that test.

That spread is the reason the two-tier structure exists. Fable’s classifiers screen or reroute some security and biology requests, which lowers its scores on tasks where the line between defensive and offensive work is thin. The company’s position, unchanged since the Fable 5 generation, is that the full capability should go first to defenders working under supervision.

Timing and the IPO

The release lands as Anthropic moves toward a public listing. The company’s IPO filing put its revenue growth and compute commitments in front of public-market investors for the first time. Several outlets framed the 5.2 launch as arriving just before the offering, and one reported that the company was aiming to list before Thanksgiving. Anthropic has not publicly confirmed a pricing date.

The timing matters because the S-1 picture depends heavily on two numbers: how fast usage grows, and how much gross margin survives that growth. A 20% list-price cut that comes with faster generation suggests the company believes its serving cost per token has fallen at least as fast as its prices. Anthropic did not break out serving costs in its launch materials.

What developers should weigh

For teams already on Fable 5.1, the move to 5.2 is a model-string change at a lower price. The 1 million token context window and the request format are unchanged, and the 5.1 cache-read discount still applies.

The harder call is whether to move work from Opus 5.5 to Fable 5.2. On Anthropic’s numbers, Fable 5.2 wins most coding and agentic benchmarks by a few points. It still costs roughly twice as much per token, and independent results on some tests put Opus 5.5 ahead or level. For high-volume pipelines, the cost per finished task, not per token, will decide the answer, and that depends on whether Fable 5.2 finishes tasks in fewer steps.

The fast mode is aimed at a different buyer. At double the standard rate for up to 2.5x the speed, it suits interactive coding tools and agent loops where latency is the bottleneck, rather than batch jobs.

What it means

This is a release about price and throughput more than new capability, and that fits where the money in frontier AI now sits. Agents that run for hours spend tokens at a rate that makes a 20% list-price cut and 30% faster output matter more than a two-point benchmark gain.

Who wins. Heavy Fable users get a straight cost cut with no migration work. Vetted security teams get a measurably stronger vulnerability-finding model in Mythos 5.2, if Anthropic’s CVE Rediscovery numbers hold up outside its own test harness. Anthropic gets a fresh product story in the weeks before it asks public investors for money.

Who is under pressure. Rival labs selling premium frontier tiers now face a lower Anthropic price point. Anthropic’s own Opus line is squeezed too: the narrower the price gap between Fable and Opus, the harder it is to explain why a buyer should stay on Opus.

What to watch next. Three things. First, whether independent evaluators converge with Anthropic’s tables, especially on SWE-bench Multimodal, where the vendor and third-party results point in opposite directions. Second, whether OpenAI and Google respond with list-price cuts on their top models rather than cache or batch discounts. Third, how the IPO roadshow handles the gap between falling token prices and the gross-margin story the company needs to tell. The two-tier Fable/Mythos model is now settled practice. The open question is how much the top tier will cost.