DeepSeek V4.1 Flash official benchmarks: Terminal-Bench 31.2 to 42.3, DeepSWE 77.0 beats Opus 5
A roundup of DeepSeek's own numbers for the Oct 17 V4.1 Flash release: MIT-licensed 763B open weights with 1M context, DeepSWE 77.0 ahead of Opus 5's 74.0, but HLE 42.6 and Terminal-Bench 42.3 still behind Opus 5.
DeepSeek called it a "minor update". Terminal-Bench went from 31.2 to 42.3. DeepSeek-V4.1-Flash-1017, out Oct 17: • Open weights, MIT license, 763B params, 1M-token context • DeepSWE v1.1: 77.0, ahead of Opus 5 (74.0) and GPT-5.6 Sol (73.0) • Agents' Last Exam: 35.4 vs 28.6 for Opus 5 • Shorter outputs and faster task runs, same API price Still well behind Opus 5 on HLE (42.6 vs 56.3) and Terminal-Bench (42.3 vs 51.8). These are DeepSeek's own numbers, so wait for independent evals. But an open "Flash" model trading blows with closed frontier models on agentic coding is the story. #AI #LLM #DeepSeek #OpenSource