V4.1 Flash 1017 comparison: 77% on DeepSWE vs Opus 5's 74% and GPT 5.6 Sol's 73%
Citing official figures, the post says V4.1 Flash 1017 hits 77% on DeepSWE with output at $0.60–$1.20 per million tokens, far below closed flagships, and discusses open-source economics.
OPEN SOURCE AI IS CLOSING THE GAP. AND THE ECONOMICS ARE GETTING HARDER TO IGNORE. DeepSeek just released V4.1 Flash 1017, an openly available, MIT licensed AI model that reportedly achieves 77% on the DeepSWE software engineering benchmark. For perspective, DeepSeek’s published comparison puts Claude Opus 5 at 74% and GPT 5.6 Sol at 73%. Now consider the economics. DeepSeek’s API costs just $0.60 to $1.20 per million output tokens, depending on usage hours. Some premium proprietary models charge tens of dollars for that same token volume. These are not identical models, and performance varies by task. But the direction of travel is becoming increasingly difficult to ignore. I’m firmly in the open source camp, and not simply because of price. It’s about competition. Independence. Innovation. The ability for researchers, developers and businesses to build powerful systems without depending entirely on a handful of corporate gatekeepers. That doesn’t mean proprietary AI has no place. It absolutely does. Companies that deliver superior performance deserve to be compensated for it. But the burden of proving that premium value is increasing. My conviction is that AI will become exponentially more valuable as intelligence becomes more accessible, not more restricted. The companies that control today’s models may not be the companies that capture tomorrow’s greatest opportunities. And that’s precisely why open source AI matters. Source: DeepSeek official release and pricing documentation. Benchmark figures are company reported.