Unverified50% confidenceBenchmarkExact time
gpt-oss-120b spent about 25 times the cost per correct answer on the distractor counting task, while its accuracy was 7.5 percentage points lower (not significant).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedGPT-6's claimed 50% price reduction refers to the API unit price, while Claude's claimed 40% reduction refers to task cost in official benchmarks; the two use different calculation methods and cannot be directly compared.73% similarUnverifiedIn a simulation of 2000 requests, the small model (1 unit cost/request) had an accuracy of 74.5% and a cost of 1.34 per correct answer; the large model (20 unit cost/request) had an accuracy of 95.8% and a cost of 20.88 per correct answer.73% similarUnverifiedOnly 38.5% of GPT-OSS 120b's capitulations adopted the decoy answer, with the remainder retreating to vagueness, providing a safer failure mode.73% similarUnverifiedDifficulty-aware routing reduces LLM calls by 85% on the final price task while closing half of the accuracy gap relative to the pure GPT-4o solution.72% similarUnverified仅用小模型策略的每正确答案成本最低($0.0004)但准确率仅33%，说明没有准确率底线的每正确答案成本指标具有误导性72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1007702API
curl https://kongchang.com/api/v1/knowledge/claims/1007702MCP
get_claim(id=1007702)