DeepSeek V4-0813 Reported as Virtually Identical to Flash Version
teortaxesTex · x · 2026-08-13
A user reported that the new DeepSeek V4-0813 is virtually identical to V4-Flash-0731 across all capabilities, and even scores lower on SciCode. The commenter suggests this is either a triumph of optimized training or a huge flop, criticizing the model's recent performance as overloaded and sloppy.
Related event: DeepSeek V4-0813 Criticized for Showing No Improvement Over Flash Version(2 posts)→
More from Models
- Elon Musk: Grok 4.7 arriving in 3-4 weeks, trained on massive SpaceX data — TinfoilTricorn · 2026-08-13
- Grok 4.6 Costs 5.4x Less Than Claude Opus 5 on Three.js Benchmark, Shattering AI Economics — SucceededMind · 2026-08-13
- Grok 4.6 Tested: Runs Autonomously for Long Periods Without Goals — XFreeze · 2026-08-13
- DeepSeek-V4-Pro Ranks #2 Among Open Models in Code Arena — arena · 2026-08-13
- Shopify CEO Tobi Reports Grok-4.6 Freaking Out Over His Low GitHub ID — johnmccrea · 2026-08-13
- MiniMax open-sources H3 video model: 33B params, tops HuggingFace, rivals closed-source — 大模型之路 · 2026-08-13