Unverified50% confidenceFactExact time
When gpt-oss is loaded with load_in_4bit = False, MXFP4 experts are retained, allowing gpt-oss-120b to be trained on a single B200
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/1/2026
First Seen
Valid until: 12/30/2026
Sources
Related Entities
Related Claims
UnverifiedParticipants in the experiment included a locally run 4-bit quantized Gemma 26B (built on MLX), Claude Opus 4.8 accessed via API, and GPT-5.4.69% similarUnverifiedGPT-OSS原生支持FP4混合精度，无需额外量化即可大幅压缩显存占用69% similarUnverifiedAdopting a one-card-one-instance approach (GPU0 running the 35B model, GPU1 running ComfyUI) rather than tensor parallelism can avoid VRAM contention when multiple clients access concurrently.68% similarUnverifiedgpt-oss-20b is able to identify the same code issues as gemma4:12B at a faster speed.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/964983API
curl https://kongchang.com/api/v1/knowledge/claims/964983MCP
get_claim(id=964983)