Unverified50% confidenceBenchmarkExact time
Based on client-side observations from two demonstrations, the author estimated an average of approximately 242 tokens per second for coding/repair tasks, approximately 269 for web tasks, and a weighted average of approximately 262 tokens per second (including reasoning tokens); factoring in request wait time, the overall rate was approximately 136 tokens per second.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/7/2026
First Seen
Valid until: 10/21/2026
Sources
Related Entities
Related Claims
UnverifiedCompleting the 30-second version involved 14 sub-agents working in coordination, with the entire process involving 779 tool calls (161 in the main conversation and 618 in sub-agents), taking approximately 2.4 hours in total.65% similarUnverifiedThe EgoDex dataset contains 338K episodes, 194 task types, and 829 hours of recorded footage61% similarUnverifiedThe entire modeling process consumed approximately 17 million input tokens and 275,000 output tokens, with a total duration of approximately 1 hour and 8 minutes.61% similarUnverifiedSlack Events API要求开发者在3秒内返回HTTP 200响应，否则Slack会启动重试机制，最多重试3次61% similarUnverifiedIn a graphic report generation test case, the model produced two files—index.html and calculations.md—with the correct total, and the complete web task took close to 6 minutes.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/984763API
curl https://kongchang.com/api/v1/knowledge/claims/984763MCP
get_claim(id=984763)