Scale Labs HiL-Bench (HiL-Bench)
We show this table for reference; we do not rank on it.
A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.
Scale score on HiL-Bench — November 12, 2026 snapshot
We mirror the published scale score view for HiL-Bench. Claude Fable 5.2 (*public set only) leads the public snapshot at 80%, followed by Muse Wonder 1.0 (79.7%) and GPT-6.1 Terra (68%). We do not use these results to rank models overall.
Claude Fable 5.2 (*public set only)
Anthropic
Muse Wonder 1.0
Meta
GPT-6.1 Terra
OpenAI
33 modelsAgenticCurrentDisplay onlyUpdated November 12, 2026 snapshot
Scale score table (33 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 (*public set only)Anthropic | 80% | Not reported | Unknown |
| 2 | Muse Wonder 1.0Meta | 79.7% | Not reported | Unknown |
| 3 | GPT-6.1 TerraOpenAI | 68% | Not reported | Closed |
| 4 | Muse Spark 1.3Meta | 66.9% | Not reported | Closed |
| 5 | MiniMax M3.1 FlashMiniMax | 65.7% | Not reported | Closed |
| 6 | GPT-6 AstraOpenAI | 65.3% | Not reported | Closed |
| 7 | GPT-6.1 SolOpenAI | 62.3% | Not reported | Closed |
| 8 | Gemini 4 ArgonGoogle | 62% | Not reported | Closed |
| 9 | Claude Fable 5.1 (*public set only)Anthropic | 61.5% | Not reported | Unknown |
| 10 | SWE-2 (*using Devin)Cognition | 61.5% | Not reported | Unknown |
| 11 | Muse Spark OpenMeta | 59.7% | Not reported | Open |
| 12 | Claude Opus 5.5Anthropic | 59% | Not reported | Closed |
| 13 | GLM-5.3Z.AI | 58.3% | Not reported | Open |
| 14 | Claude Opus 5Anthropic | 57% | Not reported | Closed |
| 15 | Claude Fable 5Anthropic | 56.3% | Not reported | Closed |
| 16 | DeepSeek V4 Flash 0731DeepSeek | 47.8% | Not reported | Open |
| 17 | GLM-5.2Z.AI | 43.7% | Not reported | Open |
| 18 | Kimi K3Moonshot AI | 42.7% | Not reported | Open |
| 19 | Claude Opus 4.7Anthropic | 41.7% | Not reported | Closed |
| 20 | Gemini 3.8 FlashGoogle | 41.5% | Not reported | Closed |
| 21 | GPT-5.5OpenAI | 39.7% | Not reported | Closed |
| 22 | DeepSeek V4 Pro 0813DeepSeek | 38.5% | Not reported | Open |
| 23 | Claude Opus 4.6Anthropic | 38.3% | Not reported | Closed |
| 24 | MiniMax M3MiniMax | 38.3% | Not reported | Open |
| 25 | Claude Opus 4.8Anthropic | 35.3% | Not reported | Closed |
| 26 | Gemini 3.1 ProGoogle | 35.3% | Not reported | Closed |
| 27 | GPT-5.6 SolOpenAI | 32.3% | Not reported | Closed |
| 28 | Gemini 3.5 FlashGoogle | 27.7% | Not reported | Closed |
| 29 | Grok 4.20 (*public set only)xAI | 20% | Not reported | Unknown |
| 30 | Kimi k2.6Moonshot AI | 18.7% | Not reported | Unknown |
| 31 | GPT-5.4OpenAI | 9.7% | Not reported | Closed |
| 32 | MiniMax M2.5MiniMax | 6.3% | Not reported | Closed |
| 33 | GPT-5.3 CodexOpenAI | 4.3% | Not reported | Closed |
How to read this leaderboard
Compare the published configurations as complete evaluation systems. The source can combine a base model, agent scaffold, tools, budget, and inference setting in each result.
Operator receipt: 33 sourced rows are currently displayable on this page; the leading published row is Claude Fable 5.2 (*public set only) at 80%.
Honest limit: This Scale table is display-only context, not benchmark provenance or a weighted model-only comparison.
How BenchLM shows HiL-Bench
BenchLM mirrors 33 published rows from Scale Labs’ public HiL-Bench leaderboard, captured on November 12, 2026 snapshot.
The table is display only. It is useful context for a published agent or model configuration, but it does not enter BenchLM’s overall or category rankings.
The published HiL-Bench snapshot places Claude Fable 5.2 (*public set only) first at 80%. The third row is 12.0 points behind. The broader top-10 range is 18.5 points, so the table still separates the published systems.
33 models have been evaluated on HiL-Bench. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. HiL-Bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About HiL-Bench
Year
2026
Tasks
33 published rows
Format
Published Scale leaderboard score
Difficulty
External agent and model evaluation
BenchLM mirrors 33 published rows from the HiL-Bench public table captured on November 12, 2026 snapshot.
Freshness and provenance
Version
HiL-Bench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does HiL-Bench measure?
A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.
Which model leads the published HiL-Bench snapshot?
Claude Fable 5.2 (*public set only) currently leads the published HiL-Bench snapshot with 80% scale score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on HiL-Bench?
The November 12, 2026 snapshot snapshot contains 33 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.