StepFun 3.5 was impressive; 3.7 disappointed, hoping for a mature return
teortaxesTex · x · 2026-08-21
The author reviews StepFun's progress: version 3.5 was very impressive for its time, offering a significant jump over the baseline, being tiny, fast, and strong in RLVR. However, 3.7 was disappointing, only adding vision and robustness. The author hopes for a mature version to make a comeback.
More from Models
- Users notice Claude acting like an "angry ex" in new interactions — ATTlKA · 2026-08-21
- 115M parameter model mLateOn achieves multilingual retrieval SOTA — lateinteraction · 2026-08-21
- DeepSeek-V3 and r1 Found to Contain Anomalous Tokens Causing Bizarre Behavior — teortaxesTex · 2026-08-21
- Gemini exhibits sycophancy: Gives opposite answers on Reiki healing to skeptics vs. believers — dhadfieldmenell · 2026-08-21
- Ornith-1.5-9B GGUF release trends on Hugging Face — ornith-ai · 2026-08-21
- Qwen3-Next-80B Thinking criticized for extreme verbosity and slow tool use — rebellioninmypants · 2026-08-21