Ranked 7 of 9 on the combined score

MiMo V2.6 Pro

MiMo Code, default reasoning. 399 timed runs on 200 tasks from 18 benchmarks.

18%Combined scoreOver the three tests
4%Within 5%64% shorter, 32% longer
4.19×Forecasting missIts typical forecast against its runtime
1.50×Retrospection missNative, its typical estimate of a finished run
48.0Benchmark scoreOut of 100, 95% interval 42.9 to 53.1

Duration following

Does it work for as long as it is asked? Each task was asked for a short and a long duration: a quarter and four times its own forecast of each task (on PaperBench and PostTrainBench, of its runtime with no time asked).

Where MiMo V2.6 Pro's runs ended, from a tenth of the time asked to ten times it, for every run and for each ask, with the timing error of each.
AskTiming errorWhere runs ended
Every timed run399 runs3.93×
Short ask200 runs2.33×
Long ask199 runs6.66×

It ran long on short asks (median 1.18× at the short ask) and stopped early on long ones (median 0.13× at the long ask). See every run

Every MiMo Pro run

Each row is one run, with a mark on the chart. Hover a row to find its mark; click a mark to find its row.

Ask
Result

399 runs

MiMo V2.6 Pro's runs. Each row links to its run page.
Loading 399 runs.

Forecasting

Before each task, it forecast how long the task would take. Each forecast is set against its own runtime on that task: its run with no time asked.

Typical miss
4.19×
Geometric mean of the factor between forecast and runtime
Most forecasts
79% too high
On 200 tasks
Median forecast
60 min
Median runtime
30 min

Retrospection

After its runs had finished, it was asked how long they took, with less time information at each step. Each figure is its typical miss against the real runtime, two answers per condition.

Oracle
1.00×
Native
1.50×
Context-only
1.67×
Replay
1.75×
Scrubbed
2.04×

Oracle adds an elapsed-time tool. Native and Context-only continue the finished session, with tools on or off. Replay sends a rebuilt transcript to the model's API, and Scrubbed does the same without timestamps or other time cues. It was stopped early, so its row covers 21 tasks.

Across the 18 benchmarks

Timing on its timed runs, and how good the work was. Score is the mean of its graded runs out of 100, from its runs with no time asked; the benchmark score is the mean of these, each benchmark counting the same.

MiMo V2.6 Pro across the 18 benchmarks
ProgramBench36 runs
10.12×
0%60.117 graded
DeepSWE28 runs
6.26×
4%50.014 graded
CORE-Bench24 runs
5.65×
0%91.712 graded
GPQA Diamond24 runs
4.67×
0%91.712 graded
TUA-Bench20 runs
4.42×
5%50.910 graded
METR public tasks4 runs
4.22×(few runs)
0%Too few graded
Agents' Last Exam24 runs
3.97×
0%57.912 graded
Humanity's Last Exam40 runs
3.91×
5%5.020 graded
BrowseComp24 runs
3.90×
0%66.712 graded
Terminal-Bench 4.032 runs
3.86×
3%6.316 graded
PPTArena20 runs
3.80×
0%39.010 graded
AppWorld28 runs
3.61×
11%35.714 graded
Sakana ALE-Bench20 runs
3.35×
0%Own units
OSWorld36 runs
3.24×
8%61.918 graded
YC-Bench8 runs
3.23×(few runs)
0%Own units
AssistantBench16 runs
2.69×
6%7.58 graded
PostTrainBench8 runs
2.48×(few runs)
25%Not scored
PaperBench7 runs
2.14×(few runs)
14%Not scored

Highest timing error first. ALE-Bench and YC-Bench keep their own units and stay out of the score, as do PaperBench and PostTrainBench.

Read before quoting. 3 runs were stopped by AgentTime, so their time is a lower bound. Details

Data and citation