ZeroBench
We show this table for reference; we do not rank on it.
A multi-step visual reasoning benchmark with pass@5 reporting and optional tool use.
Benchmark score on ZeroBench — October 29, 2026
We compile the ZeroBench rows from provider self-reports and secondary reports. Muse Spark Open leads the table at 50.0%, followed by GPT-5.4 (41.0%) and Kimi K3.1 Pro (39.0%). We do not use these results to rank models overall.
Muse Spark Open
Meta
GPT-5.4
OpenAI
Kimi K3.1 Pro
Moonshot AI
10 modelsMultimodal & GroundedCurrentDisplay onlyUpdated October 29, 2026
Benchmark score table (10 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Spark OpenMeta | 50.0% | Not reported | Open |
| 2 | GPT-5.4OpenAI | 41.0% | Not reported | Closed |
| 3 | Kimi K3.1 ProMoonshot AI | 39.0% | Not reported | Pending |
| 4 | Muse SparkMeta | 33.0% | Not reported | Closed |
| 5 | Gemini 3.1 ProGoogle | 29.0% | Not reported | Closed |
| 6 | Qwen3.8 MaxAlibaba | 24.0% | Not reported | Open |
| 7 | Kimi K3Moonshot AI | 23.0% | Not reported | Open |
| 8 | dots3-note PreviewDots Studio | 19.0% | Not reported | Open |
| 9 | Seed 2.1 ProByteDance | 18.0% | Not reported | Closed |
| 10 | Seed 2.1 TurboByteDance | 11.0% | Not reported | Closed |
Among the reported ZeroBench rows, Muse Spark Open is first at 50.0%. The third row is 11.0 points behind. The broader top-10 range is 39.0 points, so the table still separates the published systems.
10 models have been evaluated on ZeroBench. The benchmark falls in the Multimodal & Grounded category. ZeroBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About ZeroBench
Year
2026
Tasks
100 visual reasoning questions
Format
Multi-step visual reasoning
Difficulty
Tool-augmented visual reasoning
Meta evaluates ZeroBench on 100 main questions and reports pass@5, using an LLM judge to compare free-form answers against references. BenchLM stores it as a display-only multimodal reasoning benchmark.
Freshness and provenance
Version
ZeroBench 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 ZeroBench measure?
A multi-step visual reasoning benchmark with pass@5 reporting and optional tool use.
Which model scores highest on ZeroBench?
Muse Spark Open by Meta currently leads with a score of 50.0% on ZeroBench.
How many models are evaluated on ZeroBench?
10 AI models have published results on ZeroBench in the BenchLM catalog.
Compare top models on ZeroBench
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.