DeepSeek V4.1 Flash benchmarks leak: beats GPT-5.6 Sol on agentic/coding tasks at $0.14/M input

ChrisGPT · x · 2026-09-10

ChrisGPT reports DeepSeek V4.1 Flash has dropped: a 552B MoE activating 8B params for input and 16B for output, allegedly beating GPT-5.6 Sol and Opus 5 on several agentic/coding benchmarks — CyberGym 88.1 vs 84.5, Automation-Bench 54.8 vs 45.8, DeepSWE 74.2 vs 73.0. Off-peak API pricing is roughly $0.14/M input and $0.56/M output. Unconfirmed officially.

Related event: Leaked DeepSeek V4.1 Benchmarks Point to New 552B Architecture(15 posts)→

Original post →

More from Models

Models channel →