🚨 DeepSeek makes another move!
Its coding test beats Claude—is the AI agent race about to change?
DeepSeek V4.1-Flash-1017 was updated today, putting the spotlight back on the AI race! 🔥
📊 Three key highlights to watch:
✅ Another breakthrough in coding DeepSWE scored 77.0, beating Claude Opus 5 at 74.0 and GPT-5.6 Sol at 73.0.
✅ Impressive agent performance The company says the new version leads in several agent benchmarks, completing tasks faster and producing more concise outputs.
✅ A striking cost advantage The average cost per DeepSWE task is about $0.20, compared with more than $10 for Claude Opus 5. If this cost gap holds up in real-world applications, the barrier to adopting AI agents could fall further!
🌐 Open source is another trump card! The model weights are available under the MIT license. With 763 billion parameters and support for a 1-million-token context window, the model gives developers more room to deploy and innovate.
⚠️ But hold off on the big conclusions: DeepSeek still trails Claude Opus 5 on benchmarks such as HLE and Terminal-Bench. A win on an individual benchmark doesn’t mean it’s superior across the board.
💡 The AI race is shifting from “whose model is strongest” to “who can complete more real-world tasks at lower cost.”
As open-source models, AI agents, and automated execution come together, what really matters isn’t just a model’s ranking, but whether it can create lasting business value.
#以太坊跌破2400USDT
$BTC

$ETH

$BNB