DeepSeek-V4-Flash-0731 launches with agent upgrades

DeepSeek released V4-Flash-0731, a 304B MoE model, with official API in public beta. It scores 82.7 on Terminal-Bench 2.1, up from 72.1 for V4 Pro Preview. At $0.10 per DeepSWE rollout, it delivers 4.8x the solves per dollar vs GPT-5.6 Luna.
How this story unfolded
4 weeks · 10 reports · 56 community posts · 66 of 70 shown
- Jul 26
- Jul 31
- Aug 1
- Aug 2
- Aug 3
- Aug 4
- Aug 5
- Aug 6
- Aug 7
- Aug 8
- Aug 9
- Aug 10
- Aug 12
- Aug 13
- Aug 14
- Aug 18
- Aug 19
- Aug 20
DeepSeek by email
Get an email when DeepSeek has news
No news that day, no email.
More stories today
- Humyn Labs builds platform to capture human experience data for robotics
- Twin1 raises $20m seed to build AI digital twins for lawyers
- US trade talks fuel worry in Canada over tech and AI rules
- Quanta podcast probes AI reasoning with Melanie Mitchell
- AI data center opposition becomes bipartisan rallying cry before midterms