Kimi K3.1 teased to adopt DSA like DeepSeek V3.2 for better inference economics
zephyr_z9 · x · 2026-10-28
A debate over Kimi K3.1's inference economics ended with a notable tease. Responding to claims that K3.1's economics were overstated, YouJiacheng countered that under DeepInfer's announced EP64 and 77k average context, both K3 and DeepSeek V4.1 Flash can reach the batch size needed for compute-bound inference—even though K3's KV Cache is 16x that of V4.1F and its activations 6x.
Asked whether Kimi could adopt DSA like V3.2, he replied "Yes. Stay tuned."—implying K3.1 will soon get spars-attention-based economics improvements (unconfirmed).
More from Infra
- EmbeddingGemma 2 runs real-time embeddings on a MacBook CPU alone: 20ms text, 100ms images — generativist · 2026-10-28
- Statistically lossless Qwen3.8-27B GGUF quantizations land on Hugging Face, analysis paywalled — Dany0 · 2026-10-28
- Nebius benchmark: 8 small GPU replicas yield 42.5% more Cosmos video per hour — shashib · 2026-10-28
- Data centers become the new front line as AI, cyber, physical and OT threats converge — ChuckDBrooks · 2026-10-28
- preowned turns macOS built-in Apple models into an OpenAI-compatible local API — pcuenq · 2026-10-28
- 2,100+ tok/s/user for DeepSeek V4.1 Flash on 8× AMD MI355X, 2.8x faster than open-source — JiaZhihao · 2026-10-28