DeepSeek launched DeepSeek-V4.1-Flash on September 10, 2026, an open-weight AI model with 552 billion parameters and a 1 million token context window. On the DeepSWE v1.1 benchmark, V4.1-Flash scored 74.2, outperforming Anthropic’s Claude Opus 5 (74.0) and OpenAI’s GPT-5.6 Sol (73.0). The model claims inference speeds above 237 tokens per second at approximately 30 times lower cost than certain frontier closed models, with pricing at $0.15 per million input tokens and $0.60 per million output tokens during off-peak hours. Despite strong performance in coding and agentic tasks, V4.1-Flash still trails on the most complex multi-step reasoning problems. Released under the MIT license on Hugging Face, it enables enterprises to self-host the model, avoiding API dependency entirely.
Source: Read the original article
