Expired70% confidenceBenchmarkExact time
Open-source models such as Meta's Llama 3.1 series, DeepSeek-V3, and Alibaba's Qwen2.5-Coder have significantly narrowed the gap with GPT-4-level models on code generation benchmarks such as HumanEval and SWE-bench.
4
Sources
70%
Confidence
Medium-term (~90 days)
Relevance
7/12/2026
First Seen
Valid until: 10/10/2026(expired)
Sources
Alibaba Bans Claude Code: The Battle Over AI Coding Tool Security and Data Sovereignty
bilibiliSynthMindX7/4/2026
Related Claims
UnverifiedOpen-source models such as Llama 3.1 (Meta), Mistral 7B/8x7B, and Qwen2.5-Coder (Alibaba Cloud) have reached practical performance levels approaching GPT-4 on code and configuration generation tasks.83% similarVerifiedMeta's Llama 3 series, Mistral, Qwen, DeepSeek, and other models have approached or even surpassed the level of early GPT-4 on multiple benchmarks and support local deployment.79% similarUnverified以Qwen、DeepSeek、Llama为代表的开源/开放权重模型阵营正快速缩小与GPT-4o、Claude等顶级闭源模型的差距77% similarUnverifiedDeepSeek-V3 and the Qwen2.5 series have achieved performance approaching GPT-4 level on code and structured data generation benchmarks (such as HumanEval and MBPP), while their API call costs are only about 1/10 of GPT-4's.76% similarUnverifiedThe narrative instruction-following capability of quantized open-source small-parameter models such as Meta's Llama 3 series and Mistral AI's Mixtral series has approached the level of early GPT-3.5.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/489133API
curl https://kongchang.com/api/v1/knowledge/claims/489133MCP
get_claim(id=489133)