3DCodeBench: Benchmarking Agentic Procedural 3D Modeling via Code (text-to-3D) (3DCodeBench (Text to 3D))
We show this table for reference; we do not rank on it.
A benchmark for vision-language model agents that write procedural code for 3D modeling software, turning a text description into an engine-ready 3D object. This page tracks the text-prompt track.
Benchmark score on 3DCodeBench (Text to 3D) — November 13, 2026
We compile the 3DCodeBench (Text to 3D) rows from provider self-reports. LongCat-2.5 leads the table at 55.4%. We do not use these results to rank models overall.
1 modelCodingCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (1 model)
The one result on this page was reported by the lab that built the model.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | LongCat-2.5MeituanReported by Meituan | 55.4% | Reported by Meituan | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
How to read this leaderboard
Operator receipt: 1 sourced row is currently displayable on this page; the leading published row is LongCat-2.5 at 55.4%.
Honest limit: Automated geometry metrics may not capture perceptual quality, and the launch row is a provider run whose scoring setup is not described.
About 3DCodeBench (Text to 3D)
Year
2026
Tasks
Text prompts to procedural 3D modeling code
Format
Benchmark score (0-100)
Difficulty
Code-level geometric reasoning and 3D software APIs
3DCodeBench (May 2026) evaluates how well models act as procedural 3D modelers by translating text and image references into modeling-software code, with an execution environment for iterative refinement and a companion 3DCodeArena for pairwise human preferences. The authors find most failures come from API mismatches, with floating or disconnected geometry in successful renders. Meituan reports LongCat-2.5 at 55.4 on the text-to-3D track in its November 2026 model card, next to in-house runs of GPT-5.5 (51.9), GPT-5.6 Sol (52.3), Claude Opus 4.8 (52.7), Claude Opus 5 (54.3), GLM-5.3 (53.4), Kimi K3 (49.3) and DeepSeek V4.1 Flash (53.3). BenchLM stores the LongCat-2.5 row; the other figures are Meituan runs listed for context.
Freshness and provenance
Version
3DCodeBench (Text to 3D) 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 3DCodeBench (Text to 3D) measure?
A benchmark for vision-language model agents that write procedural code for 3D modeling software, turning a text description into an engine-ready 3D object. This page tracks the text-prompt track.
Which model scores highest on 3DCodeBench (Text to 3D)?
LongCat-2.5 by Meituan currently leads with a score of 55.4% on 3DCodeBench (Text to 3D).
How many models are evaluated on 3DCodeBench (Text to 3D)?
1 AI models have published results on 3DCodeBench (Text to 3D) 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 7,000+ subscribers.
One email each week. Unsubscribe anytime.