Taste-Bench
Taste-Bench was published by The Decoder GPT-5.6 Sol was the best performing model on Taste-Bench GPT-5.5 was the second best performing model on Taste-Bench Claude was tested on Taste-Bench Grok was tested on Taste-Bench DeepSeek was tested on Taste-Bench GLM was tested on Taste-Bench MiniMax was tested on Taste-Bench Mistral was tested on Taste-Bench SWE-bench Pro provided data for Taste-Bench METR's public transcript release MALT provided data for Taste-Bench Qwen3.6-27B was used as a base model for Taste-Bench training Indra runs 15× faster than Tesseract OCR on CPU Datalab released Indra, a 1M-parameter OCR model· Graphiti 生成的实体摘要
关系变化
按生效时间- GPT-5.6 SolBEST MODEL ONTaste-Bench
GPT-5.6 Sol was the best performing model on Taste-Bench
- METR's public transcript release MALTPROVIDED DATA FORTaste-Bench
METR's public transcript release MALT provided data for Taste-Bench
- CollabBenchTESTED ONTaste-Bench
MiniMax was tested on Taste-Bench
- ClaudeTESTED ONTaste-Bench
Claude was tested on Taste-Bench
- SWE-benchPROVIDED DATA FORTaste-Bench
SWE-bench Pro provided data for Taste-Bench
- DeepseekTESTED ONTaste-Bench
DeepSeek was tested on Taste-Bench
- Grok 4.7TESTED ONTaste-Bench
Grok was tested on Taste-Bench
- Ray‑BanTESTED ONTaste-Bench
GLM was tested on Taste-Bench
- Taste-BenchPUBLISHED BYTHE DECODER
Taste-Bench was published by The Decoder
- MistralTESTED ONTaste-Bench
Mistral was tested on Taste-Bench
- TeraWulfUSED AS BASE MODEL FORTaste-Bench
Qwen3.6-27B was used as a base model for Taste-Bench training
- GPT-5.5SECOND BEST MODEL ONTaste-Bench
GPT-5.5 was the second best performing model on Taste-Bench
- Taste-BenchOUTPERFORMSTasteVal
Indra runs 15× faster than Tesseract OCR on CPU
- High Taste Testers Episode OneRELEASEDTaste-Bench
Datalab released Indra, a 1M-parameter OCR model
关系是 Graphiti 从报道中提取的事实陈述，不是原文引述；失效时间表示后来的报道显示这条关系已不成立。