Unverified50% confidenceBenchmarkExact time
Claude Haiku 4.5, Gemma 4 31B, and Qwen3-235B verbatim copied the coordinates provided by users, and two other models added a prohibited "points" field on GPS checkpoint requests.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/11/2026
First Seen
Valid until: 1/9/2027
Sources
Related Entities
Related Claims
UnverifiedThe nine checkpoints consist of four target models (Qwen3's 4B, 8B, 14B, and Gemma 4's 12B) multiplied by three algorithms68% similarUnverifiedThe dataset covers four open-weight models — Qwen2.5-7B, Mistral-7B-v0.3, Phi-3.5-mini, and Qwen2.5-1.5B — with 30 trajectories each.64% similarUnverifiedThe author noted that other models previously tested, such as 35B and 3B variants, actually performed worse than the Q2 version of Qwen3 27B on the Kanban task.63% similarUnverifiedThe "Qwen3.8-27B" in the title is likely a memory error or typo by the poster, as the commonly seen Qwen specifications in the community are 32B, 72B, etc.61% similarUnverifiedMapping all three tiers—Sonnet, Opus, and Haiku—to the same model endpoint is required for a domestic model to fully take over Claude Code; omitting Haiku will cause errors or fallback to the default configuration on lightweight requests.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999530API
curl https://kongchang.com/api/v1/knowledge/claims/999530MCP
get_claim(id=999530)