Expired50% confidenceBenchmarkExact time
In the Agency tool-calling test, both models scored 95 points, and the only failure for both was task No. 13 (USD to JPY foreign currency conversion).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/9/2026
First Seen
Valid until: 10/7/2026(expired)
Sources
16GB VRAM Showdown: Ornith 35B vs Qwen 35B — Full Benchmark Comparison
bilibili程序员-智能译站7/7/2026
Related Claims
UnverifiedDevin 2.0 resolved 13.86% of real-world programming problems in the SWE-bench benchmark, compared to only 1.96% for previous AI models.67% similarUnverifiedIn a needle-in-a-haystack test with a 256K context, both models performed identically, each achieving a score of 93%, with each failing only once at 100% full context depth.63% similarUnverified在PlanBench的神秘方块世界测试中，2023年的模型几乎完全失败，但O1等推理模型的出现带来了显著改善61% similarUnverifiedOpenAI's chart shows that at lower test-time compute levels, the model's success rate on this problem is close to zero, with occasional success only under extremely long reasoning chains.61% similarUnverified测试数据表明RNN头与马尔可夫头的草稿通过率差距仅在1-2%以内，而部署成本差距可达数倍60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/342894API
curl https://kongchang.com/api/v1/knowledge/claims/342894MCP
get_claim(id=342894)