MiMo-V2.6-Flash vs MiMo-V2.6-Pro

Overall
45
vs
Overall
48

Too close to call. MiMo-V2.6-Pro wins 23 of the 42 test cases where they differ.

Learn to build evals like this in the AI Engineering Academy

Task by task

A score turns orange only when that model wins clearly more of the task's test cases. Under each score, the average time per test case on the machine it ran on. MiMo-V2.6-Pro spends 1.3× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
451m 21s
4.2K, 92% thinking5.9K, 89% thinking
482m 13s
Agents
381m 19s
2.2K, 86% thinking8.7K, 84% thinking
474m 48s
Structured Extraction
6238.0s
2.6K, 91% thinking3.4K, 87% thinking
601m 26s
Coding
242m 40s
9.9K, 97% thinking10.7K, 95% thinking
242m 58s
Grounded Answers
6242.1s
2.6K, 89% thinking3.2K, 89% thinking
621m 9s
Instruction Following
641m 2s
4.8K, 90% thinking4.5K, 86% thinking
591m 23s
Business Logic
4356.4s
4.4K, 96% thinking7.1K, 92% thinking
571m 52s
Security & permissions
662m 6s
3.8K, 88% thinking3.6K, 86% thinking
711m 41s
Diagnosis & investigation
01m 33s
1.5K, 83% thinking6.2K, 86% thinking
02m 54s

What you'd run

The build behind each score. Orange marks the faster, smaller, cheaper or longer-context side.

SpecAB
Speed
66 tok/s (better)
55 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1.1M
1.1M
Cost per run
$0.33 (better)
$1.80
Cost per case
$0.0021 (better)
$0.012
Time per case
1m 21s (better)
2m 13s
Quantization
BF16
BF16
Harness
openrouter
openrouter

The race

7 lead changes

Each model's running score as the test cases go by. The dashed line is a perfect run.

MiMo-V2.6-FlashMiMo-V2.6-Pro
025507510050100150Perfect run4548

Every test case

153 test cases

Each square is one test case, colored by the model that scored higher.

19 won by MiMo-V2.6-Flash23 won by MiMo-V2.6-Pro111 tied