Anthropic launches Claude Fable 5.2, Mythos 5.2, cuts API prices 20%

Anthropic has released two new frontier language models — Claude Fable 5.2 for general use and Claude Mythos 5.2 for cybersecurity work — and says both set new high scores across a range of third-party benchmarks while cutting API prices by 20%, according to VentureBeat. For anyone choosing between rival models, the more useful finding may be buried in the launch paperwork: the same model's cost and quality swing widely depending on how hard you push it to think.

What did Anthropic announce, and what does it cost?

According to VentureBeat, Fable 5.2 is the general-purpose model, while Mythos 5.2 uses the same underlying model with different safeguards and is reserved for vetted cybersecurity and life sciences users in Anthropic's Project Glasswing program. Both now cost $8 per million uncached input tokens and $40 per million output tokens, down from $10 and $50 for the 5.1 generation — the 20% cut. Cached reads fall to $0.20, and Anthropic's pricing table lists fast-mode rates of $16 input and $80 output. The cheaper Opus 5.5 still undercuts them at $4 and $20. Fable is available now on paid Claude plans, in Claude Code and through the Claude Platform (model ID claude-fable-5-2), as well as on AWS, Google Cloud and Microsoft Azure.

How do the new Claude models compare with rivals?

Anthropic's launch comparison, as reported by VentureBeat, has one of the two new models leading on 13 of 15 reported score lines. The clearest gap is in coding: on FrontierCode 2.0, which asks whether a patch is good enough to merge into a real open-source repository, Fable scored 49.1%, ahead of GPT-6 Astra at 42.8% and Google's Gemini 4 Carbon at 41.1% — though Fable ran at Xhigh effort and Carbon at Medium. Fable also led FrontierSWE v2 with 69.9%, versus 65.5% for Astra and 61.2% for Carbon. Carbon, which Google announced on October 28 and still limits to selected defenders through its Fairwind program, took the business-automation test, 53.9% to Fable's 51.4%. Opus 5.5 edged Fable on the AA-Briefcase head-to-head ranking, 1,800 to 1,785. Mythos leads the reported cyber and protein-design results, though Anthropic's system card notes that labs used different tasks for CyScenarioBench, making that a looser comparison with GPT-6 Cyber.

Why does the effort setting change the bill so much?

This is the part most likely to affect a budget. On FrontierCode 2.0's 132 tasks, Fable's score rose from 37.6% at Medium to 45.3% at High and 49.1% at Xhigh, while the system card's cost breakdown puts those settings at $2.48, $6.72 and $14.19 per task. At High, Fable cost almost exactly what Astra spent at Max ($6.68) for a score 2.5 points higher, and roughly half of Opus at Max ($12.82). Push Fable to Max and the average bill climbs to about $44.70 — while the score falls below what it reached at Xhigh. Because Medium is the default in Claude apps and Claude Code, the headline result requires a deliberate change. Against the 5.1 generation, Fable 5.2 at Medium beat the older model's best (32.4% at Xhigh, $8.68 per task) by 5.2 points at about 71% lower cost; at Xhigh it gained 16.7 points but cost about 63% more per task despite the lower list price.

What does this mean for coding and knowledge-work teams?

Anthropic itself recommends Opus 5.5 for most coding and knowledge work, positioning Fable for jobs that already burn time and money on retries — a migration across an unfamiliar codebase, or a bug that survives several rounds of diagnosis — or as an advisor a cheaper model consults when it gets stuck, according to VentureBeat. The launch testimonials, which Anthropic selected, point to measurable signals: Vercel engineer Joe Haddad said projects that took weeks with other models were finished in days, and Datadog's Matt Lee compared the model to an incident commander that sends faster models to gather information. The benchmarks reinforce a tune-by-workload approach. On the offline OSWorld 2.1 test, Fable completed every required step on 54.6% of tasks versus 53.2% for Opus; on AA-Briefcase it led the rubric pass rate, 66.1% to 61.2%, even as Opus won the head-to-head ranking. On BenchCAD 2.1 with tools, Fable reached 83.5 at Max versus 65.3 for Opus. For workflows that write to a CRM, the system card suggests tracking the share of jobs that land every required update and how often a human has to repair partial changes.

How does this connect to Grafy?

Grafy runs many cloud models side by side, which makes the story's practical lesson easy to act on: compare models on your own tasks, then match effort to the job. In Grafy Chat you can put the same prompt to different models and read the answers next to each other. For the coding benchmarks that dominate the launch, Grafy Code is the studio where model choice and effort settings matter most. And the lowest-cost routing approach in Grafy mirrors Anthropic's core finding — spend the cheapest model that can answer, and escalate to a stronger one only when a task stalls. Whether or not a given frontier model is in your workspace yet, that habit of measuring cost per completed task, rather than per token, is the transferable idea.

Frequently asked questions

Did Anthropic cut API prices for both new models?

Yes. According to VentureBeat, both Fable 5.2 and Mythos 5.2 now cost $8 per million uncached input tokens and $40 per million output tokens, down from $10 and $50 for the 5.1 generation — a 20% reduction. Cached reads dropped to $0.20. For comparison, Anthropic's cheaper Opus 5.5 is listed at $4 input and $20 output, so the new frontier models still cost more per token than that smaller option.

Does the higher effort setting always cost more?

Yes, per-task cost rises with each step up. On FrontierCode 2.0, Fable's average cost climbed from $2.48 at Medium to $6.72 at High and $14.19 at Xhigh, and to about $44.70 at Max, according to Anthropic's system card. But the top setting scored below Xhigh, so paying the most did not buy the best result — a reminder to match effort to the task rather than always maxing it out.

Is Mythos 5.2 a different model from Fable 5.2?

Not fundamentally. According to VentureBeat, Mythos uses the same underlying model as Fable but with different safeguards, and Anthropic reserves it for vetted cybersecurity and life sciences users in its Project Glasswing program. Fable 5.2, by contrast, is available broadly — on paid Claude plans, in Claude Code and through the Claude Platform, plus AWS, Google Cloud and Microsoft Azure — so most developers will encounter Fable rather than Mythos.

---

Source: VentureBeat — AI — Anthropic releases Claude Fable 5.2 and Mythos 5.2, pairing new, record-high benchmarks with 20% API cost reduction