This week's briefing (9–15 Nov)Data updated 6:35 IST, 22 NovHow we check claims
Liang Wenfeng
DeepSeek · Lab leaders. 2 sourced entries, newest first. Curated coverage with primary sources, not a complete feed; a role establishes context, not authority.
DeepSeek · 17 October 2026 · Company statement
DeepSeek ships V4.1-Flash-1017 update: 'wish-driven programming', open weights and $0.75/1k Search API
Primary source, read in fullModelsOpen weightsAI agents
DeepSeek's official news post describes the V4.1-Flash-1017 update as a Pareto improvement — stronger benchmarks, faster task completion and shorter outputs — trained with reinforcement learning on self-feedback for end-to-end project building (3D modeling, web games, game-engine recreation). It announces open-source weights on Hugging Face and ModelScope and standalone Search & Fetch APIs priced at $0.75 per 1,000 calls.
What this could change
Editorial: an open-weight model marketed on finishing whole projects from a one-line wish, plus a dirt-cheap search API, hands indie builders capabilities that were agency deliverables a year ago.
Perspective & interests
DeepSeek under Liang Wenfeng monetizes API usage and mindshare from open weights, so framing 1017 as a cost-performance Pareto win drives both.
Reviewed 2026-11-06. Entry ID liang-wenfeng-deepseek-v41-flash-1017-2026-10 · Permalink
DeepSeek releases V4.1-Flash: 552B MoE with asymmetric encoder-decoder and halved off-peak pricing
Primary source, read in fullModelsInference costs
DeepSeek's official API news post announces DeepSeek-V4.1-Flash, a 552B-parameter MoE with a new causal encoder-decoder design using only 8B active parameters for input and 16B for output, which it says beats DeepSeek-V4-Pro on benchmarks. The post cites KV-cache cuts to 1/4 the HBM and 1/8 the SSD storage of the previous generation, native visual understanding, and new API pricing effective 10 September 2026 with off-peak rates at 50% of peak.
What this could change
Editorial: DeepSeek keeps pushing frontier-adjacent capability at aggressively low inference prices, which resets the cost floor builders should expect to pay for API tokens.
Perspective & interests
Liang Wenfeng's DeepSeek competes on price-performance, so efficiency claims for V4.1-Flash are the core of its API business pitch.
Reviewed 2026-11-06. Entry ID liang-wenfeng-deepseek-v41-flash-2026-09 · Permalink