DeepSeek V4.1 Flash beats frontier models on agent benchmarks
DeepSeek released V4.1 Flash 1017, a fast, low-cost model with a 1M-token context and an MIT license. It scored 42.3 on Terminal-Bench 4.0 versus 39.9 for GPT-5.6 Sol and 77.0 on DeepSWE v1.1 versus 74.0 for Opus 5, while trailing on HLE without tools (42.6 vs 56.3).
- Terminal-Bench 4.0: 42.3 for V4.1 Flash vs 39.9 for GPT-5.6 Sol
- DeepSWE v1.1: 77.0 vs 74.0 for Opus 5 and 73.0 for GPT-5.6 Sol
- HLE without tools: 42.6 vs 56.3 for Opus 5
- 1M-token context, up to 256K output, MIT license
Read next
AI