Overview
DeepSeek is a Chinese AI laboratory that has gained prominence for its efficient frontier models and significant fundraising efforts. The lab has been working to retain researchers amid intense competition from rivals whose valuations have soared.
Recent Developments
-
2026-07-31: Official deepseek-v4-flash 0731 public beta — same 284B/13B MoE + 1M context; re-post-training + dspark; Responses API / Codex; MIT weights (2026-08-01-deepseek-v4-flash-0731-official-release)
-
2026-07-25: Verbally pauses second fundraising round after viral posts attributed to liang-wenfeng investor-meeting remarks; first ~71B pre-money (2026-07-26-deepseek-halts-second-fundraising-round)
-
See also deepseek-second-fundraising-pause and fundraising-leak-risk
-
2026-07-16: Included among providers in meta-oversight-board LLM free-expression study (2026-07-16-meta-oversight-board-llm-free-expression-assessment); inkling sized against DeepSeek V4 class opens
-
2026-04-24: Released DeepSeek-V4 with 1.6 trillion parameters (Pro) and 1 million token context window
-
2026-04-24: DeepSeek-V4-Pro achieves 80.6% on SWE-bench Verified (comparable to Claude Opus 4.6 and GPT-5.4)
-
2026-04-24: DeepSeek-V4-Flash (284B parameters, 13B active) achieves 79.0% on SWE-bench Verified
-
2026-04-24: Open-source MIT License, weights on Hugging Face
-
2026-04-24: Pricing ~$3.48/M output tokens (about 1/7th cost of competitors)
-
2026-04-24: Hybrid attention architecture (CSA + HCA) reduces KV cache by 90%, inference FLOPs by 73%
-
2026-04-23: Reportedly fundraising at $20 billion valuation to retain researchers
Contradiction
April 2026 press cited a ~7B June round and a verbally paused second round aimed near ~20B as early/unconfirmed relative to later round figures.
Business Context
- Talent competition: AI researchers increasingly mobile between labs
- Valuation trends: Frontier AI labs commanding premium valuations
- China ecosystem: Chinese AI labs navigating both domestic and international dynamics
- Open-source focus: DeepSeek positioning as accessible alternative to closed models
Related
-
ai-startups: Broader startup funding context
-
ai-investment-trends: Market trends affecting valuations
-
frontier-ai-labs: Competitive landscape
-
us-china-ai-competition: Geopolitical context
-
open-source-ai: Open-source AI movement
-
model-efficiency: Efficiency innovations