JobBench
An occupational agent benchmark for professional workflows that workers say they most want delegated to AI.
Top models on JobBench — October 29, 2026
As of October 29, 2026, Muse Wonder Full leads the JobBench leaderboard with 75.3% , followed by Muse Wonder (70.5%) and Muse Spark 1.3 (64.9%).
Muse Wonder Full
Meta
Muse Wonder
Meta
Muse Spark 1.3
Meta
35 modelsAgentic6% of Agentic reference weightCurrentUpdated October 29, 2026
Leaderboard (35 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Wonder FullMeta | 75.3% | Not reported | Closed |
| 2 | Muse WonderMeta | 70.5% | Not reported | Closed |
| 3 | Muse Spark 1.3Meta | 64.9% | Not reported | Closed |
| 4 | Ultramarine 1 FlashPaper Instruments | 62.9% | Not reported | Closed |
| 5 | MiMo-V2.6-ProXiaomi | 62.0% | Not reported | Open |
| 6 | Muse Spark OpenMeta | 61.8% | Not reported | Open |
| 7 | Hy4 previewTencent | 61.7% | Not reported | Open |
| 8 | MiMo-V2.6-FlashXiaomi | 61.2% | Not reported | Open |
| 9 | Step 5 PreviewStepFun | 59.0% | Not reported | Open |
| 10 | Qwen3.8-Flash-NextAlibaba | 55.7% | Not reported | Open |
| 11 | Muse Spark 1.1Meta | 54.7% | Not reported | Closed |
| 12 | Qwen3.8 MaxAlibaba | 53.4% | Not reported | Open |
| 13 | Kimi K3Moonshot AI | 52.9% | Not reported | Open |
| 14 | Atria Dawn PreviewShanghai Artificial Intelligence Laboratory | 50.3% | Not reported | Open |
| 15 | Claude Opus 4.7 (Adaptive)Anthropic | 45.9% | Not reported | Closed |
| 16 | GPT-5.5OpenAI | 42.7% | Not reported | Closed |
| 17 | GPT-5.4OpenAI | 38.9% | Not reported | Closed |
| 18 | Claude Sonnet 4.6Anthropic | 36.9% | Not reported | Closed |
| 19 | Claude Opus 4.6Anthropic | 36.7% | Not reported | Closed |
| 20 | GPT-5.2OpenAI | 34.3% | Not reported | Closed |
| 21 | GPT-5.3 CodexOpenAI | 33.7% | Not reported | Closed |
| 22 | Qwen3.8-27BAlibaba | 33.4% | Not reported | Open |
| 23 | Claude Opus 4.5Anthropic | 32.3% | Not reported | Closed |
| 24 | Claude Sonnet 4.5Anthropic | 27.7% | Not reported | Closed |
| 25 | GPT-5.1-CodexOpenAI | 26.2% | Not reported | Closed |
| 26 | GPT-5.2-CodexOpenAI | 26.0% | Not reported | Closed |
| 27 | Claude 4.1 OpusAnthropic | 21.9% | Not reported | Closed |
| 28 | Qwen3.5 PlusAlibaba | 18.5% | Not reported | Closed |
| 29 | Claude 4 SonnetAnthropic | 18.4% | Not reported | Closed |
| 30 | Claude Haiku 4.5Anthropic | 16.0% | Not reported | Closed |
| 31 | MiniMax M2.5MiniMax | 14.8% | Not reported | Closed |
| 32 | Gemini 3 ProGoogle | 11.4% | Not reported | Closed |
| 33 | Gemini 3 FlashGoogle | 11.4% | Not reported | Closed |
| 34 | Kimi K2.5Moonshot AI | 8.7% | Not reported | Open |
| 35 | GPT-5 (high)OpenAI | 8.5% | Not reported | Closed |
According to BenchLM.ai, Muse Wonder Full leads the JobBench benchmark with a score of 75.3%, followed by Muse Wonder (70.5%) and Muse Spark 1.3 (64.9%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
35 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.
How many models are evaluated on JobBench?
35 AI models have published results on JobBench in the BenchLM catalog.
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 6,100+ subscribers.
One email each week. Unsubscribe anytime.