Thursday, September 10, 2026

DeepSeek-V4.1-Flash ships.

After Wednesday Issue 13 close, DeepSeek-V4.1-Flash shipped with new API id deepseek-flash — Causal Encoder–Decoder MoE (vendor: 552B; 8B active input / 16B output), native multimodal, 1M ctx / 384K max out, MIT open weights on Hugging Face, and new peak/off-peak pricing effective 04:00 UTC Sep 10. Legacy V4 Flash / Vision-Exp route to V4.1-Flash; deepseek-v4-pro routes to Flash from Sep 14 04:00 UTC until V4.1-Pro (FLAG vendor evals). LMSYS Day-0 SGLang+Miles serving methodology for V4.1 (FLAG project benches; SGLang still 0.5.19). Gemini Enterprise Computer Use + Shell sandboxes GA (Sep 9) with VPC-SC/PSC/CMEK/pause-resume. Ollama v0.34.0 GA — ChatGPT Desktop via the macOS Ollama app + Apple Silicon structured-output note (no official numbers).

Window after Issue 13 close (~2026-09-09 morning ET) through 2026-09-10 morning. Primary sources only. Vendor evals flagged.