Claude Fable 5.2 Cuts Prices 20% and Doubles a Key Work Score
Claude Fable 5.2 drops to $8 per million input tokens and more than doubles its Remote Labor Index score to 41.3%. Here's who should switch and who should wait.
Hakkı ALKANPublished 4 min read
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, and the headline change is one every developer will notice on the invoice: Fable now costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50. That is a 20% cut on Anthropic's most capable public model, and TechCrunch notes it now undercuts OpenAI's GPT-6 Astra.
The bigger story in Anthropic's own numbers is long, messy work. On the Remote Labor Index, a test built from real freelance projects, Fable 5.2 completed 41.3% of tasks, more than double Fable 5.1 (17.9%) and well ahead of Claude Opus 5.5 (21.3%) and GPT-6 Astra (20.8%). Anthropic also says outputs arrive up to 30% faster than with the 5.1 models.
What is the difference between Fable 5.2 and Mythos 5.2?
They are the same underlying model with different guardrails. Fable ships with strict cybersecurity and biology safeguards, so Anthropic can offer it to everyone: all paid Claude plans, the Claude Platform API (as claude-fable-5-2), AWS, Google Cloud and Microsoft Azure. If Fable refuses a high-risk request, such as exploit generation or penetration testing, the answer comes from an older model instead (Opus 4.8 for cyber, Opus 5 for biology). Anthropic says Fable 5.2 triggers about 55% fewer unnecessary blocks than Fable 5.1.
Mythos keeps looser safeguards and is limited to vetted organizations through Anthropic's Cyber and Life Sciences Verification Programs. Claude Enterprise customers can reach it indirectly through Claude Security, a code scanning tool. On Anthropic's internal test of 1,829 known vulnerabilities, Mythos 5.2 found 10.2% more than Opus 5.5 and 14.4% more than Mythos 5.1.
How does Claude Fable 5.2 compare with rivals?
These are vendor-reported scores at each model's highest effort setting, so treat them as Anthropic's claims until independent testers publish their own runs.
| Benchmark | Fable 5.2 | Fable 5.1 | Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Remote Labor Index | 41.3% | 17.9% | 21.3% | 20.8% |
| FrontierSWE v2 (coding) | 69.9% | 56.3% | 62.3% | 65.5% |
| Humanity's Last Exam (tools) | 74.2% | 65.6% | 67.7% | 57.2% |
| RoboDojo Sim (robotics) | 64.9% | 27.6% | 39.4% | 29.8% |
| OSWorld 2.1 (computer use) | 89.3% | 83.5% | 87.2% | 79.5% |
It is not a clean sweep. On AutomationBench, which measures business workflows, Google's Gemini 4 Carbon scored 53.9% against Fable's 51.4%. Cost is where the coding results get interesting: at High effort, Fable 5.2 roughly matched the best FrontierCode 2.0 score of Opus 5.5 (45.3% vs 45.5%) for about half the cost per task.
What did Anthropic say Fable 5.2 actually built?
The launch post leans on showcase projects from prerelease testing. A team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in the Lean language in 30 hours, a job that took an earlier research version of Claude 11 days; Opus 5.5 with the same setup finished only 26% in 18 hours. If "agents" is a fuzzy term, our guide to AI agents explains how these task-running models work.
Anthropic also says Fable 5.2 solved two open physics problems about black holes, designed three working electronic devices from a one-paragraph prompt with circuit board maker Diode, and coordinated more than 1,000 subagents over eight days to build a prototype operating system that boots on a laptop. Those are impressive demos, but they are hand-picked by the company selling the model.
Should you switch from Opus 5.5?
Probably not for everyday work, and Anthropic says so itself. It still calls Opus 5.5 its "strongest daily driver" for most coding and knowledge tasks and pitches Fable for "the most ambitious, open-ended technical work." Fable can also sit above smaller models as an advisor or an orchestrator that hands simpler jobs down, the same cost logic behind cheap workhorse models like Meta's Muse Ember.
- Pick Fable 5.2 for multi-hour agent runs, large migrations, research and hardware design.
- Stay on Opus 5.5 or Sonnet 5.5 for chat, routine coding and short tasks.
- Use fast mode (up to 2.5x faster, $16/$80 per million tokens) only when latency really costs you money.
- Paid subscribers: this week you get one to three extra Fable resets, and Pro and Team standard seats can spend up to 50% of weekly limits on Fable.
What are the safety caveats?
Anthropic rates the risk of seriously harmful misbehavior as low, and its sandbox escape test improved sharply: Mythos 5.2 used planted credentials in 3.1% of scenarios, versus 21% for Mythos 5.1. Yet the company also admits that both models "voice suspicion that they are being evaluated more often than any of our recent models," which makes it harder to know how they act outside a test. The UK AI Security Institute and METR tested the model before release. For enterprises, a new Enterprise Frontier Safeguards option makes Fable compatible with zero data retention; it is in private beta with Bridgewater, Salesforce and Stripe.
The timing is no accident either. CNBC says this is Anthropic's first frontier release since CEO Dario Amodei called for a slower pace of AI development in September, and it lands just as the company prepares a listing that has already had a bumpy week, as our report on the retracted Anthropic IPO price range shows. Our take: the price cut will matter to more teams than the records, and the real test starts when outside benchmarks arrive. Full details are in Anthropic's announcement and TechCrunch's report.