DeepSeek V4.1 Flash 1017 reportedly scores 42.3 on Terminal-Bench 4.0, beating GLM-5.3 but only with its own harness
teortaxesTex · x · 2026-10-19
X user teortaxesTex says he's confident DeepSeek V4.1 Flash 1017's 42.3 score on Terminal-Bench 4.0 is real — above GLM-5.3 — and asked Artificial Analysis to rerun it in the native harness. The quoted post reveals a key caveat: without DeepSeek's own harness the model scores only 32, jumping to 40+ with dsh. The leaked table also shows clear gains over V4.1-Flash on agentic benchmarks like DeepSWE v1.1 (77.0), Automation-Bench (60.5) and Apex Bench (72.1), though HLE still trails Claude Opus 5 (56.3). The harness dependence is worth watching.
Related event: DeepSeek V4.1 Flash benchmark score disputed by third-party tests(2 posts)→
More from Models
- 763B DeepSeek-V4.1-Flash runs locally at 500+ tok/s on 4x RTX Pro 6000 — TheZachMueller · 2026-10-19
- DeepSeek opens Search API: tester spends $0.08 on 110 calls, finds authority scores tied to hosts — gaganghotra_ · 2026-10-19
- DeepSeek-V4.1-Flash-1017 shown building full projects from scratch — FinanceYF5 · 2026-10-19
- MiniMax teases MiniMax-M3.1-Flash in official comeback announcement — MiniMax_AI · 2026-10-19
- Viral theory claims RSI-capable models can't be open-sourced, DeepSeek dates are frozen snapshots — teortaxesTex · 2026-10-19
- Claude Is Today's Busiest Bridge in AI Discourse Graph as Anthropic Brand Buzz Cools 17% — YvesMulkers · 2026-10-19