GPT-6 Astra posts best-ever Blender 3D modeling result in dev's homemade benchmark
repligate · x · 2026-09-05
Developer SkyeSharkie ran GPT-6 Astra through the first task of a Blender 3D modeling benchmark they are building, and was stunned by the result.
- Multi-turn one-shot modeling of a drill from a SINGLE reference image angle, in a no-image-gen, text-only test
- Best result seen from any model so far, beating fable 5.1 and 5.6 sol
- Author's take: 3D modeling is only a few generations from being completely solved by LLMs
- Prompt guidance for the benchmark is still unrefined; full results to be published later
Note: GPT-6 Astra access is unverified and results are day-one, rough-eval numbers.
Related event: GPT-6 Astra's 3D modeling demos flood in, community calls it SOTA(37 posts)→
More from Models
- Opus 5.5 tops OpenRouter spending in 20 of 29 task categories — mmurph · 2026-10-21
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21