Fetching from the wire…
01
02
03
04
05
06
07
08
Infra2026-08-21 · source-backed
On a warmed M3 Ultra A/B, Qwen3.8-27B went from 458 tok/s baseline at 4K prefill to 588 with the ANE/GPU split and 625 with ANE/CPU/GPU, costing about 7 GB peak memory. GitHub The compile-memory drop is arguably the bigger deal, since a 35.8 GB spike locked out most Macs from ever building the banks. Same release adds an M2 Ultra simdgroup_matrix indexer kernel for DeepSeek-V4-Flash, 1.37x to 1.38x faster from 32K to 1M context while staying bit-exact.
Each link below shares sources, entities, or timing with this story.
Unsloth Studio supports DeepSeek / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (Unsloth Studio supports DeepSeek); both cover CPU, DeepSeek, Flash, GPU; reported by the same outlet (github.com).
Ollama supports DeepSeek / Shared entities / Earlier coverage / Tension
Linked by a graph relationship (Ollama supports DeepSeek); both cover GitHub, GPU, Macs, Qwen3; earlier GitHub coverage from 2026-04-23.
GPT competes with DeepSeek / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (GPT competes with DeepSeek); both cover DeepSeek, Flash, Qwen3, Same; reported by the same outlet (github.com).
Ollama supports DeepSeek / Shared entities / Earlier coverage
Linked by a graph relationship (Ollama supports DeepSeek); both cover DeepSeek, Flash, Qwen3; earlier DeepSeek coverage from 2026-08-10.
DeepSeek released DeepSeek V4 Pro / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (DeepSeek released DeepSeek V4 Pro); both cover GitHub, Qwen3; reported by the same outlet (github.com).
openclaude uses DeepSeek / Shared entities / Same source domain / Earlier coverage / Tension
Linked by a graph relationship (openclaude uses DeepSeek); both cover DeepSeek, GitHub, GPU; reported by the same outlet (github.com).
DeepSeek competes with Anthropic / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (DeepSeek competes with Anthropic); both cover CPU, GPU; reported by the same outlet (github.com).
Linked by a graph relationship (DeepSeek competes with Anthropic); both cover GitHub, Same; reported by the same outlet (github.com).