Seonglae ChoA bigger model that actually cuts long-context inference cost significantly. The key is KV cache reduction: 3,514 bytes/token in V4-Flash → 890 bytes/token, roughly a 4x drop.
- Causal Encoder-Decoder: only about half the model runs during prefill, reducing input processing cost.
- SWA + Bounded Replay: local KV isn't stored; only the recent portion is recomputed when needed.
- CSA2: multiple layers share long-term KV and retrieval results.
- Hierarchical Sparse Indexer: instead of searching all 1M tokens at every layer, candidates are narrowed down progressively.
- Remaining KV is stored in FP4.
As a result, at 1M context KV is roughly 890MB vs 3.5GB before — about 1/4 the HBM and 1/8 the SSD cache. This is especially advantageous for long-running agents.
DeepSeek V4.1-Flash-1017
The October 17 update uses reinforcement learning with self-feedback for end-to-end project development. DeepSeek reports faster task completion and shorter outputs than the previous version. Search and Fetch APIs each cost $0.75 per 1,000 calls; server-side tool calls for DeepSeek API users begin billing on October 31, 2026, at 16:00 UTC.
Lucebox DeepSeek V4.1-Flash-1017 quantizations
These builds target the Lucebox engine on AMD Radeon AI PRO R9700 and Ryzen AI Max. Routed experts use ROCMFP low-bit formats, while attention, shared experts, and the output head use Q8_0. The model card's three base recipes occupy 383–405 GB and report 78.6–81.5% top-1 agreement with the teacher, illustrating the storage–quality tradeoff rather than full-quality equivalence.