Anthropic Claude Fable 5.2 at launch
- Anthropic
- Released November 9, 2026
- 1M context
Anthropic released Claude Fable 5.2 on November 9, 2026. As of November 14, 2026, it leads the LLMBenchmark Index of 118 models, ahead of Google Gemini 4 Carbon. Anthropic's own announcement reports 11 benchmark results, including Humanity's Last Exam (74.2%) and AutomationBench (51.4%). It costs $8.00 per million input tokens and $40.00 per million output tokens on OpenRouter.
What Anthropic reported at launch
From Anthropic's announcement, November 9, 2026. The maker's own numbers, not re-run by us.
-
AA-Briefcase v1.1
1785
-
AA-Briefcase v1.1
66.1%
rubric
-
AutomationBench
51.4%
#3 of 20 models with a reported result
-
Chartography
73.4
no tools
#2 of 13 models with a reported result
-
Humanity's Last Exam
74.2%
with tools
#1 of 38 models with a reported result
-
FrontierCode 2.0
49.1%
Xhigh effort
#1 of 4 models with a reported result
-
FrontierSWE v2
69.9%
#1 of 5 models with a reported result
-
OSWorld 2.1 (offline set, partial)
89.3%
#1 of 2 models with a reported result
-
Remote Labor Index (RLI)
41.3%
High effort
#1 of 3 models with a reported result
-
RoboDojo Sim
64.9%
#1 of 3 models with a reported result
-
WANDR (relaxed score)
82.1%
#1 of 3 models with a reported result
Independent results so far
68 scores from 68 boards that test models themselves. Ranks compare Claude Fable 5.2 with every model on the same board.
Agentic work (15)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| BioMysteryBench (Vals AI) | 83.7% | #1 of 29 | — | Vals AI boards |
| Design Arena: agents building full-stack apps | 1402 | #1 of 50 | — | Design Arena |
| Design Arena: spreadsheets | 1434 | #1 of 39 | — | Design Arena |
| LiveBench long agentic | 82.9% | #1 of 31 | — | LiveBench |
| LiveBench research | 94.5% | #1 of 31 | — | LiveBench |
| Remote Labor Index | 41.3% | #1 of 16 | — | Epoch AI benchmarking hub |
| Design Arena: agents building mobile apps | 1325 | #2 of 55 | — | Design Arena |
| Terminal-Bench-Science (Vals AI) | 55.7% | #2 of 41 | Mini-SWE-agent | Vals AI boards |
Show all 15
| Vals Science Index | 63.7% | #2 of 23 | — | Vals AI boards |
|---|---|---|---|---|
| APEX-Agents | 80.1% | #3 of 59 | — | Epoch AI benchmarking hub |
| SpireClimb | 15.5% | #4 of 9 | Max | Epoch AI benchmarking hub |
| LiveBench browser | 86.9% | #7 of 31 | — | LiveBench |
| Legal Research Bench (Vals AI) | 46.1% | #18 of 83 | — | Vals AI boards |
| Tax Agent Bench (Vals AI) | 32.1% | #25 of 77 | — | Vals AI boards |
| Harvey's Legal Agent Benchmark (Vals AI) | 5.0% | #28 of 34 | — | Vals AI boards |
Coding (15)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| FrontierCode 2.0 | 49.1% | #1 of 55 | High | Artificial Analysis |
| FrontierSWE | 69.9% | #1 of 22 | Max | Epoch AI benchmarking hub |
| GameBench 2 | 99.1 | #1 of 89 | xHigh | DuelLab GameBench 2 |
| LMArena WebDev | 1838 | #1 of 105 | xHigh | LMArena WebDev |
| LMArena WebDev: React | 1915 | #1 of 97 | xHigh | LMArena (Hugging Face dataset) |
| ProgramBench (Vals AI) | 19.0% | #1 of 61 | — | Vals AI boards |
| SciCode | 67.7% | #1 of 109 | Max | Artificial Analysis |
| Terminal-Bench 4 (Vals AI) | 65.2% | #1 of 57 | Mini-SWE-agent | Vals AI boards |
Show all 15
| Vibe Code Bench (Vals AI) | 93.8% | #1 of 105 | OpenHands | Vals AI boards |
|---|---|---|---|---|
| CursorBench 4.0 | 60.0% | #2 of 20 | max | CursorBench (Cursor) |
| SWE-bench Multimodal (Vals AI) | 62.3% | #2 of 20 | — | Vals AI boards |
| ALE-Bench | 2509.8 | #4 of 110 | High | ALE-Bench (Sakana AI) |
| Terminal-Bench 4.0 (Artificial Analysis run) | 61.6% | #4 of 111 | xHigh | Artificial Analysis |
| LMArena WebDev: HTML | 1738 | #7 of 104 | xHigh | LMArena (Hugging Face dataset) |
| LiveBench coding | 79.2% | #12 of 31 | — | LiveBench |
Reasoning (14)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| ARC-AGI | 98.5% | #1 of 75 | xHigh | Epoch AI benchmarking hub |
| Chess puzzles | 79.0% | #1 of 81 | Max | Epoch AI benchmarking hub |
| CritPt v2, hard subset (Artificial Analysis run) | 76.5% | #1 of 56 | Max | Artificial Analysis |
| EBR-bench | 90.5% | #1 of 29 | Max | Epoch AI benchmarking hub |
| GoBench | 2566 | #1 of 19 | High | GoBench (Roland Gao) |
| HLE-Diamond (Artificial Analysis run) | 80.2% | #1 of 74 | Max | Artificial Analysis |
| Humanity's Last Exam (Artificial Analysis run) | 74.3% | #1 of 187 | Max | Artificial Analysis |
| Humanity's Last Exam (Scale run) | 72.4% | #1 of 19 | — | Scale SEAL leaderboards |
Show all 14
| Mystery game puzzles | 90.0% | #1 of 75 | Max | Epoch AI benchmarking hub |
|---|---|---|---|---|
| SimpleBench | 94.9% | #1 of 72 | — | Epoch AI benchmarking hub |
| MysteryMechanism (Vals AI) | 51.8% | #2 of 28 | — | Vals AI boards |
| ARC-AGI-2 | 93.8% | #4 of 76 | xHigh | Epoch AI benchmarking hub |
| GPQA Diamond (Epoch AI run) | 93.9% | #14 of 115 | Max | Epoch AI benchmarking hub |
| LiveBench reasoning | 88.8% | #15 of 31 | — | LiveBench |
General (7)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| DTBench | 99.2% | #1 of 130 | — | Epoch AI benchmarking hub |
| Furniture assembly | 90.0% | #1 of 36 | Max | Epoch AI benchmarking hub |
| LMCA | 70.6% | #1 of 125 | xHigh | Epoch AI benchmarking hub |
| Vals Index | 71.3% | #1 of 34 | — | Vals Index |
| GDPval-AA | 1831 | #4 of 117 | Max | Artificial Analysis |
| LiveBench data analysis | 81.5% | #7 of 31 | — | LiveBench |
| Long context reasoning (AA-LCR) | 83.3% | #14 of 176 | Max | Artificial Analysis |
Web design (6)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| LiveBench design | 85.1% | #1 of 31 | — | LiveBench |
| Design Arena: 3D | 1517 | #2 of 121 | — | Design Arena |
| Design Arena: websites | 1351 | #3 of 134 | — | Design Arena |
| Design Arena: all code categories | 1373 | #4 of 129 | — | Design Arena |
| Design Arena: UI components | 1371 | #5 of 124 | — | Design Arena |
| Design Arena: game dev | 1396 | #6 of 128 | — | Design Arena |
Math (5)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| FrontierMath, tier 4 | 100.0% | #1 of 66 | Max | Epoch AI benchmarking hub |
| FrontierMath, tiers 1 to 3 | 95.1% | #1 of 80 | Max | Epoch AI benchmarking hub |
| ProofBench | 100.0% | #1 of 75 | Max | Epoch AI benchmarking hub |
| MathArena | 82.8% | #2 of 72 | max | MathArena |
| OTIS Mock AIME | 99.7% | #13 of 115 | Max | Epoch AI benchmarking hub |
Knowledge (4)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| AA-Omniscience | 66.8 | #1 of 58 | Max | Artificial Analysis |
| SimpleQA Verified | 94.0% | #1 of 77 | Max | Epoch AI benchmarking hub |
| PRBench Legal | 49.0% | #13 of 39 | — | Scale SEAL leaderboards |
| PRBench Finance | 47.1% | #17 of 33 | — | Scale SEAL leaderboards |
Intelligence (2)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| AA Intelligence Index | 64.6 | #1 of 55 | — | Artificial Analysis (via OpenRouter) |
| Epoch Capabilities Index | 173.4 | #1 of 125 | — | Epoch AI benchmarking hub |