Unverified50% confidenceBenchmarkExact time
According to a UC Berkeley test, seven frontier models showed no significant differences in SWE-Bench Lite scores across different harnesses such as Claude Code and Codex, and the newer the model, the smaller the differences.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/22/2026
First Seen
Valid until: 1/20/2027
Sources
Former Cursor Engineer: Is Agent Harness Dead? The Data Flywheel and Three Things the Model Can't Absorb
bilibili全球AI科技不费眼10/22/2026
Related Entities
Related Claims
UnverifiedOn a contamination-free benchmark, the trained 8B model surpassed Claude Opus 4.8 and GPT-6 Astra on the creative utility metric.62% similarUnverifiedThe blogger pointed out four discount factors in the benchmark comparison: different test frameworks (TerminalBench 1017 ran within the ClaudeCode framework while the baseline ran within the DeepSeek framework), Artificial Analysis's own test of the old Flash on TerminalBench scored only 26.8% rather than 31.2%, the competitor comparison used Opus 5 and GPT 5.6 Sol rather than the latest models, and the old model's NL2 repo score is inconsistent across different tables (64.0 vs 65.4).61% similarUnverifiedThe T0 tier contains only two tools, Claude Code and Codex, both with a composite score of 8.4, tied for first place.58% similarUnverifiedMistral 7B证明精心训练的7B模型可以匹敌早期的13B甚至30B模型58% similarUnverifiedThe performance degradation caused by lost formatting cues is reproducible across six answer models, covering five models across the Claude Opus, Sonnet, and Haiku tiers plus OpenAI's GPT-5.6.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1087768API
curl https://kongchang.com/api/v1/knowledge/claims/1087768MCP
get_claim(id=1087768)