Unverified50% confidenceFactExact time
There is currently no public benchmark data showing the performance of LFM3 and d1 on MLX.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/27/2026
First Seen
Valid until: 11/10/2026
Sources
Related Entities
Related Claims
UnverifiedThe commit message for this version contains 'mlx: add LFM3 and d1 models (#19333)', meaning the core change is adding the LFM3 and d1 models to the MLX backend.80% similarUnverifiedThe LFM3 technical details officially disclosed so far are limited, and which modalities are specifically covered and the performance level of each modality still await empirical verification.75% similarUnverifiedThe capability foundation in this update was extracted from commit 36d46a0c3 on the system_one_mlx branch, with the MLX scoring logic and manifest-list-related changes split out to be progressed separately.63% similarUnverifiedLFM2.5-24B-A2B is a pure-text reasoning model with no vision tower, always begins with a <think> block, and those needing image capabilities should use LFM2.5-VL-3B (the highest model in the LFM2.5 VL series).62% similarUnverifiedvLLM's `fp8` KV cache format corresponds to E4M3 and offers three calibration methods: no calibration (scaling factor defaults to 1.0), random-token warmup, and dataset calibration via llm-compressor.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1105273API
curl https://kongchang.com/api/v1/knowledge/claims/1105273MCP
get_claim(id=1105273)