Claude Fable 5.2 released: the new League leader costs 20% less than its predecessor.
Claude Fable 5.2 was released on November 9, 2026 and enters the AI Pro League at #1 with a League Score of 100.0 — built from a LiveBench-topping 84.9 — at $8/$40 per million tokens, a 20% price cut from Claude Fable 5.1. This post prices the release on a real workload, shows the normalisation maths behind the perfect score, and states plainly what a single-benchmark record does and does not prove.
By AI Pro League · Published

Claude Fable 5.2 is the new leader of the AI model rankings. Our November 10, 2026 changelog records the release from Anthropic’s official pricing page and model documentation (platform.claude.com, read today): released Nov 9, 2026 at $8.00 input / $40.00 output per million tokens, cached input $0.20 — a 20% cut from Claude Fable 5.1’s $10.00 / $50.00. Its first recorded benchmark is the field’s best: 84.9 on LiveBench 2.0 (results dated 2026-11-06), the new #1 on that leaderboard. What follows is computed from the database at build time.
Claude Fable 5.2: the release record, priced and dated
| Measure | Fable 5.2 | Fable 5.1 |
|---|---|---|
| Release date | Nov 9, 2026 | Sep 1, 2026 |
| Input / output per 1M tokens | $8.00 / $40.00 | $10.00 / $50.00 |
| Cached input per 1M tokens | $0.20 | $0.25 |
| LiveBench raw score | 84.9 | 84.0 |
| League Score / overall rank | 100.0 / #1 | 96.0 / #5 |
| Score coverage | 40% | 55% |
| Context window / max output | 1,000,000 / 128,000 tokens | 1,000,000 / 128,000 tokens |
The shape of the release is unusual: the newer model is cheaper on every rate — input, output and cached input all fell by exactly 20% — while the context window (1,000,000 tokens), the max output (128,000 tokens), the June 2026 knowledge cutoff and the text+image inputs are unchanged from Fable 5.1. Its gated twin, Claude Mythos 5.2 — the same model behind a verification program — is not tracked, consistent with how this database handles gated specialist models. The full rate card sits in our AI model pricing index.
How 84.9 becomes a League Score of 100.0
Fable 5.2’s League Score rests on this single component, so the perfect score is checkable by hand. Normalisation stretches the recorded LiveBench field — 34 models, raw scores from 32.4 to 84.9, median 68.9 — across a 0–100 scale, and Fable 5.2 is the field maximum: (84.9 − 32.4) ÷ 52.5 × 100 = 100.0. With one recorded component at 40% weight, the League Score is that normalised value: 100.0. Our LiveBench explainer sets out why this benchmark carries the joint-largest weight of the five components, and the LiveBench leaderboard carries the full field.
What the 20% price cut is worth on a real workload
| Model | League Score | Workload cost |
|---|---|---|
| Claude Fable 5.2 | 100.0 | $160.00 |
| Claude Fable 5.1 | 96.0 | $200.00 |
| GPT-6 Astra | 99.4 | $200.00 |
| Claude Opus 5.5 | 97.1 | $80.00 |
Against its own predecessor the saving is $40.00 on the standard workload — the 20% cut applied to a $200.00 bill. Cached input deepens it: at $0.20 per million, cached tokens cost 97.5% less than fresh input, so a cached-heavy run of 8M cached + 2M fresh input + 2M output costs $97.60 instead of $160.00 uncached — the pattern priced across the field in our prompt caching pricing analysis. The honest counterweight sits in the same table: Claude Opus 5.5 scores 97.1 — within three points of the top — at exactly half Fable 5.2’s workload cost.
The top of the League now
| Rank | Model | League Score | Coverage |
|---|---|---|---|
| 1 | Claude Fable 5.2 (new) | 100.0 | 40% |
| 2 | GPT-6 Astra | 99.4 | 55% |
| 3 | Gemini 4 Carbon | 97.3 | 40% |
| 4 | Claude Opus 5.5 | 97.1 | 40% |
| 5 | Claude Fable 5.1 | 96.0 | 55% |
| 6 | Claude Opus 5 | 95.6 | 90% |
Fable 5.2 displaces GPT-6 Astra at #1 by 0.6 points, and pushes Fable 5.1 from the top spot it held in our League table price-gap analysis down to #5. Anthropic now holds three of the top five places.
What the #1 ranking does not yet prove
- The record is one benchmark deep. Coverage is 40% — LiveBench only. No GPQA Diamond, SWE-bench, LMArena or MMLU-Pro result is recorded yet, so the 100.0 says “best on the freshest general benchmark we track”, not “best at everything” — the standing caveat in our score coverage explainer.
- The predecessor’s record is deeper. Fable 5.1 carries a second component — a maker-reported GPQA Diamond 93.7 — at 55% coverage. If Fable 5.2 posts weaker science or coding results, its score will move when they land; a single-component 100.0 has nowhere to hide.
- The gap at the top is small. The top four models sit within four points of each other, and LiveBench re-measurements have moved this table before — today’s changelog alone revised more than twenty LiveBench rows under LiveBench 2.0.
None of that dims the release itself: a new flagship that tops the freshest leaderboard we track while cutting every price by a fifth is the strongest launch pattern this database has recorded. The score to watch is the second one — whichever benchmark reports on Fable 5.2 next will test whether 100.0 is a ceiling or a starting point.