Anthropic Ships Claude Fable 5.2 at $8/$40, 20% Below Fable 5.1
Fable 5.2 is live for all users at $8/$40 per million tokens, and Anthropic reports it ahead of Opus 5.5 on agentic coding. On the arena's landing-page brief it was cheaper than Fable 5.1 at low effort but cost about twice as much at high, because it wrote nearly four times the tokens.
Anthropic released Claude Fable 5.2 today, alongside Claude Mythos 5.2. For anyone paying per token, the headline is the rate card: $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1, a 20% cut.
- Available to all users today on the Claude Platform as claude-fable-5-2, and on Amazon Web Services, Google Cloud and Microsoft Azure.
- Fast mode costs double the base rate, $16/$80 per million tokens, for up to 2.5x the generation speed.
- Anthropic says output is up to 30% faster than Fable 5.1.
- Effort levels run low, medium, high, xhigh and max. Medium is the default in the Claude apps and Claude Code.
Mythos 5.2 is the same underlying model without certain cybersecurity and biology safeguards. It is limited to vetted organizations, which apply through Anthropic's Cyber Verification and Life Sciences Verification programs.
Where it lands on the benchmarks
Every score below is vendor-reported, from Anthropic's announcement, with each model at its highest available effort unless noted.
| Benchmark | Fable 5.2 | Fable 5.1 | Opus 5.5 |
|---|---|---|---|
| FrontierCode 2.0 | 49.1% (xhigh) | 32.4% (xhigh) | 45.5% |
| FrontierSWE v2 | 69.9% | 56.3% | 62.3% |
| OSWorld 2.1 (offline set, partial) | 89.3% | 83.5% | 87.2% |
| Remote Labor Index | 41.3% (high) | 17.9% | 21.3% |
| Humanity's Last Exam (with tools) | 74.2% | 65.6% | 67.7% |
| AutomationBench | 51.4% | 31.4% | 42.5% |
Anthropic pairs the scores with a cost claim: on FrontierCode 2.0 at high effort, Fable 5.2 scores 45.3%, about the same as Opus 5.5's best of 45.5%, for "roughly half the cost per task". It does not lead every row. Gemini 4 Carbon tops AutomationBench at 53.9% against 51.4%, and Opus 5.5 edges it on the AA-Briefcase v1.1 Elo, 1800 to 1785.
It works until the job is actually complete instead of stopping at a plausible checkpoint - Andrew Mager, Staff Engineer, Agentic Systems, Uber
Safeguards: about 55% fewer blocks
Anthropic says the cyber and biology safeguards are as robust as those of Fable 5.1 but produce fewer unnecessary blocks. In early testing across claude.ai, Claude Code and Cowork, Fable 5.2 had about 55% fewer blocks than Fable 5.1: about 60% fewer in cybersecurity and 40% fewer in biology. A blocked query is answered by another model instead: Opus 4.8 for cybersecurity, Opus 5 for biology.
First arena run: low to max
Fable 5.2 ran on launch day against the arena's landing-page brief, the same one-shot, single-file brief every model on this site gets, through the Claude Code CLI in headless mode with an explicit effort setting per run. Cost is the API list-price equivalent the CLI reports for each run; output tokens include thinking.
| Effort | Cost | Output tokens | Of which thinking | Wall clock | Page size |
|---|---|---|---|---|---|
| low | $1.35 | 33,600 | 3,144 | 5.1 min | 64 KB |
| medium | $2.08 | 51,466 | 12,250 | 8.0 min | 82 KB |
| high | $4.72 | 117,729 | 60,854 | 18.6 min | 119 KB |
| xhigh | $9.07 | 175,230 | 124,618 | 27.8 min | 104 KB |
| max | $9.09 | 226,508 | 193,411 | 37.6 min | 68 KB |
Fable 5.1 ran the same brief at the same effort labels. Each cell reads Fable 5.1 → Fable 5.2:
| Effort | Cost | Output tokens | Wall clock |
|---|---|---|---|
| low | $2.17 → $1.35 | 27,403 → 33,600 | 4.9 → 5.1 min |
| medium | $2.10 → $2.08 | 26,521 → 51,466 | 4.9 → 8.0 min |
| high | $2.25 → $4.72 | 29,690 → 117,729 | 5.4 → 18.6 min |
| xhigh | $5.58 → $9.07 | 95,653 → 175,230 | 18.0 → 27.8 min |
| max | $10.12 → $9.09 | 144,332 → 226,508 | 30.3 → 37.6 min |
Fable 5.2 came out cheaper per run at low and at max, but not because of the price cut. At low it cost $1.35 against $2.17 on 23% more tokens; at list rates the output tokens alone come to $1.34 for Fable 5.2 and $1.37 for Fable 5.1, so the gap sits on the input side of the bill. Max is the same story: $9.09 against $10.12, yet its output tokens alone cost more, $9.06 against $7.22, and the Fable 5.1 run carried about $2.90 of input-side cost. Medium cost the same, $2.08 against $2.10, on nearly twice the tokens.
At high and xhigh the order flips. High wrote nearly four times the tokens of Fable 5.1 high and cost 2.1x as much; xhigh cost 1.6x. A 20% cheaper token does not offset four times as many. Within the Fable 5.2 ladder, thinking is what grows: it is 9% of output tokens at low, 24% at medium, 52% at high, 71% at xhigh and 85% at max.
Per token, it was faster. Measured as output tokens per second of wall clock, Fable 5.2 ran at 100 to 111 against 79 to 94 for Fable 5.1, 16 to 26% faster at all five labels and inside Anthropic's "up to 30%". The longer runs come from the token count, not the speed. One brief and one run per effort level make this a sample, not a verdict.
The benchmark table is vendor-reported; the cost, token and time figures are the arena's own measurements. None of these numbers says which landing page is better, and that is what the blind vote is for. Put Fable 5.2 high against Fable 5.1 high on the landing brief and pick one. Runs on the other top briefs (game, dashboard, tool, news, chip8 and more) are landing tonight, and the Fable 5.2 page shows each one as it finishes.