DeepSeek ships V4.1-Flash-1017: Terminal-Bench jumps 31.2→42.3, MIT-licensed
petrusenko_max · x · 2026-10-23
DeepSeek quietly shipped V4.1-Flash-1017, framed as a minor update to V4.1-flash but with big gains (self-reported): Terminal-Bench 4.0 from 31.2 to 42.3, ProgramBench from 20.3 to 29.1, HLE from 36.8 to 42.6. On DeepSeek's own table it beats GPT-5.6 Sol on DeepSWE (77.0 vs 73.0) and Terminal-Bench (42.3 vs 39.9), while Opus 5 still leads most rows (51.8 on Terminal-Bench). The company also says the model finishes tasks faster with shorter outputs. Open weights, MIT license; independent evals pending.
Related event: DeepSeek Open-Sources V4.1-Flash-1017 with Big Terminal-Bench Gains(2 posts)→
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23