Unverified50% confidenceBenchmarkExact time
After raising llm-retry's max_attempts to 8, all 100 calls passed, but it took 16.6 seconds, more than double the theoretical lower bound, and sent 375 requests.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the long-context archiving experiment, 8 requests each input approximately 261,000+ tokens and each completed approximately 1,022 outputs; the total throughput of the joint decoding window was 214.85 tokens per second, approximately 26.86 per request, but the window was only approximately 2.28 seconds and prefilling was serial or interleaved.78% similarUnverifiedThe client-side rate limiter allocates send time slots according to the server's replenishment rate, and in the test achieved all 100 succeeding, 100 requests, 0 429s, and the last answer completing at 8.0 seconds.76% similarUnverifiedAfter honoring retry-after and layering full jitter on top, the last call completed at 11.8 seconds, 313 requests were sent, and p95 was 9.1 seconds, with every metric better than the two backoff formulas.73% similarUnverifiedThe retry loop in the code attempts at most 4 times, with exponential backoff starting at 500 milliseconds, and it prioritizes obeying the Retry-After response header when the service provides one.73% similarUnverifiedUnder a 4M TPM premise, long Agent turns with a 400K context can trigger at most 10 requests per minute (hitting the TPM cap), and full-context requests only 4 per minute.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1064738API
curl https://kongchang.com/api/v1/knowledge/claims/1064738MCP
get_claim(id=1064738)