Head to head

Any two models from the board, on one scale.

Model A

Model B

openrouter · deepseekhosted

overall

◂ 6

openrouter · qwenhosted

60vs54

DeepSeek V4.1 Flash leads by 6 overall · ahead on 7 of 9 tasks

tasks

Task by task

9 tasks

78TOOL+1 ▸79

52STRUCT◂ +547

63RAG◂ +2241

100CTX◂ +1189

41CODE+34 ▸75

39REASON◂ +732

46INSTR◂ +2125

55CLASS◂ +1738

70SUM◂ +1060

best build per side · same suite, same graders · hover a score for cases passed

specs

215 tok/sspeed64 tok/s

—memory—

1.0Mcontext1M

$0.7017cost / run$0.5705

—quantization—

openrouterharnessopenrouter

the lit value is the better spec — faster, smaller, cheaper, or more context · quantization & harness just describe the builds