JobBench
An occupational agent benchmark for professional workflows that workers say they most want delegated to AI.
Top models on JobBench — November 11, 2026
As of November 11, 2026, Muse Wonder Full leads the JobBench leaderboard with 75.3%, as reported by Meta , followed by Inkling 2 Small (70.5%, as reported by Thinking Machines Lab) and Muse Wonder (70.5%, as reported by Meta).
Muse Wonder Full
Meta
Reported by MetaInkling 2 Small
Thinking Machines Lab
Reported by Thinking Machines LabMuse Wonder
Meta
Reported by Meta37 modelsAgentic6% of Agentic reference weightCurrentUpdated November 11, 2026
Leaderboard (37 models)
Of 37 results: 20 benchmark-run, 17 lab-reported.
Ranks on this table count the 20 benchmark-run and independent results. A lab-reported result keeps its place by score, without a rank.
- Lab-reported
- 17
- Independent
- 0
- Benchmark-run
- 20
Showing all 37 results, in score order.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| No rank on this table | Muse Wonder FullMetaReported by Meta | 75.3% | Reported by Meta | Not reported | Closed |
| No rank on this table | Inkling 2 SmallThinking Machines LabReported by Thinking Machines Lab | 70.5% | Reported by Thinking Machines Lab | Not reported | Closed |
| No rank on this table | Muse WonderMetaReported by Meta | 70.5% | Reported by Meta | Not reported | Closed |
| No rank on this table | LongCat-2.5MeituanReported by Meituan | 65.4% | Reported by Meituan | Not reported | Open |
| No rank on this table | Muse Spark 1.3MetaReported by Meta | 64.9% | Reported by Meta | Not reported | Closed |
| No rank on this table | Ultramarine 1 FlashPaper InstrumentsReported by Paper Instruments | 62.9% | Reported by Paper Instruments | Not reported | Closed |
| No rank on this table | MiMo-V2.6-ProXiaomiReported by Xiaomi | 62.0% | Reported by Xiaomi | Not reported | Open |
| No rank on this table | Muse Spark OpenMetaReported by Meta | 61.8% | Reported by Meta | Not reported | Open |
| No rank on this table | Hy4 previewTencentReported by Tencent | 61.7% | Reported by Tencent | Not reported | Open |
| No rank on this table | MiMo-V2.6-FlashXiaomiReported by Xiaomi | 61.2% | Reported by Xiaomi | Not reported | Open |
| No rank on this table | Step 5 PreviewStepFunReported by StepFun | 59.0% | Reported by StepFun | Not reported | Open |
| No rank on this table | Qwen3.8-Flash-NextAlibabaReported by Alibaba | 55.7% | Reported by Alibaba | Not reported | Open |
| No rank on this table | Muse Spark 1.1MetaReported by Meta | 54.7% | Reported by Meta | Not reported | Closed |
| No rank on this table | Qwen3.8 MaxAlibabaReported by Alibaba | 53.4% | Reported by Alibaba | Not reported | Open |
| No rank on this table | Kimi K3Moonshot AIReported by Moonshot AI | 52.9% | Reported by Moonshot AI | Not reported | Open |
| No rank on this table | Atria Dawn PreviewShanghai Artificial Intelligence LaboratoryReported by the lab | 50.3% | Reported by the lab | Not reported | Open |
| 1 | Claude Opus 4.7 (Adaptive)AnthropicRun by JobBench | 45.9% | Run by JobBench | Not reported | Closed |
| 2 | GPT-5.5OpenAIRun by JobBench | 42.7% | Run by JobBench | Not reported | Closed |
| 3 | GPT-5.4OpenAIRun by JobBench | 38.9% | Run by JobBench | Not reported | Closed |
| 4 | Claude Sonnet 4.6AnthropicRun by JobBench | 36.9% | Run by JobBench | Not reported | Closed |
| 5 | Claude Opus 4.6AnthropicRun by JobBench | 36.7% | Run by JobBench | Not reported | Closed |
| 6 | GPT-5.2OpenAIRun by JobBench | 34.3% | Run by JobBench | Not reported | Closed |
| 7 | GPT-5.3 CodexOpenAIRun by JobBench | 33.7% | Run by JobBench | Not reported | Closed |
| No rank on this table | Qwen3.8-27BAlibabaReported by Alibaba | 33.4% | Reported by Alibaba | Not reported | Open |
| 8 | Claude Opus 4.5AnthropicRun by JobBench | 32.3% | Run by JobBench | Not reported | Closed |
| 9 | Claude Sonnet 4.5AnthropicRun by JobBench | 27.7% | Run by JobBench | Not reported | Closed |
| 10 | GPT-5.1-CodexOpenAIRun by JobBench | 26.2% | Run by JobBench | Not reported | Closed |
| 11 | GPT-5.2-CodexOpenAIRun by JobBench | 26.0% | Run by JobBench | Not reported | Closed |
| 12 | Claude 4.1 OpusAnthropicRun by JobBench | 21.9% | Run by JobBench | Not reported | Closed |
| 13 | Qwen3.5 PlusAlibabaRun by JobBench | 18.5% | Run by JobBench | Not reported | Closed |
| 14 | Claude 4 SonnetAnthropicRun by JobBench | 18.4% | Run by JobBench | Not reported | Closed |
| 15 | Claude Haiku 4.5AnthropicRun by JobBench | 16.0% | Run by JobBench | Not reported | Closed |
| 16 | MiniMax M2.5MiniMaxRun by JobBench | 14.8% | Run by JobBench | Not reported | Closed |
| 17 | Gemini 3 ProGoogleRun by JobBench | 11.4% | Run by JobBench | Not reported | Closed |
| 18 | Gemini 3 FlashGoogleRun by JobBench | 11.4% | Run by JobBench | Not reported | Closed |
| 19 | Kimi K2.5Moonshot AIRun by JobBench | 8.7% | Run by JobBench | Not reported | Open |
| 20 | GPT-5 (high)OpenAIRun by JobBench | 8.5% | Run by JobBench | Not reported | Closed |
A dash in the rank column marks a lab-reported result. It keeps its place by score, without a rank on this table.
- Lab-reported
- Benchmark-run
- Independent
According to BenchLM.ai, Muse Wonder Full leads the JobBench benchmark with a score of 75.3%, as reported by Meta, followed by Inkling 2 Small (70.5%, as reported by Thinking Machines Lab) and Muse Wonder (70.5%, as reported by Meta). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
37 models have been evaluated on JobBench. The benchmark falls in the Agentic category. BenchAlign v5.8 gives JobBench 6% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About JobBench
Year
2026
Tasks
130 tasks across 35 occupations
Format
Agentic workplace deliverables
Difficulty
Professional multi-source workflows
JobBench evaluates agents on source-grounded workplace tasks across 35 occupations, using heterogeneous reference files and fact-anchored chained rubrics. BenchLM tracks the main-set score.
Freshness and provenance
Version
JobBench 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 JobBench measure?
An occupational agent benchmark for professional workflows that workers say they most want delegated to AI.
Which model scores highest on JobBench?
Muse Wonder Full by Meta currently leads with a score of 75.3% on JobBench, as reported by Meta.
How many models are evaluated on JobBench?
37 AI models have published results on JobBench in the BenchLM catalog.
Compare top models on JobBench
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.