Unverified50% confidenceBenchmarkExact time
Code-type tasks can achieve a throughput improvement of about 4x or more (154 vs 18-19 tok/s), while prose-type tasks achieve only about 2x (35-37 vs 18-19 tok/s).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/25/2026
First Seen
Valid until: 1/23/2027
Sources
Related Entities
Related Claims
UnverifiedIn code scenarios, each verification step accepts about 7.4 tokens, reaching a throughput of 154 tok/s; in prose scenarios each step accepts about 2.5 tokens, with a throughput of 35-37 tok/s; the ordinary decoding baseline throughput is about 18-19 tok/s.72% similarUnverifiedIn the 16 benchmarks officially published for 2.5-24, it scored 27.69 on automated tasks, with second place at 8.72, leading by about 3.2 times.68% similarUnverifiedAfter 200 steps, the maximum discrepancy between the corresponding weights of the two runs reaches 0.237.65% similarUnverifiedOn the 35 prompts that the router kept local, Llama 3.2 1B's answers were judged acceptable 30 times (86%, average 4.4/5); on the 23 text prompts sent to the cloud, the same model was acceptable only 4 times (17%, average 2.0/5).65% similarUnverifiedThe Program Bench score rose from 20.3 to 29.1, an improvement of 8.8 points, using the 'Almost' criterion (passing 5 of 9 tests counts as a pass).64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1096606API
curl https://kongchang.com/api/v1/knowledge/claims/1096606MCP
get_claim(id=1096606)