skynet · reasoning
SKYNET

minimax-m2.7

Skynet prior-gen

⚖ COMPARE MODELS
VALS INDEX
58.0% +/- 22.0
Accuracy
VALS INDEX
26.0s
Latency (median)
$VALS INDEX
$0.1176
Cost / Test (1P-est)
SCORE STATUSscored
TEST SETS13
RECORDED TRIALS700
CUSTOM PASS64%
DEEP VALIDATION · 2026-08-20

Dual-judge family campaign

FULL REPORT →
CONSENSUS MEAN3.867 / 5
BOTH-JUDGE PASS60.0%
CROSS-JUDGED45 / 45
MEDIAN WALL21.18s
GPQA_DIAMONDserialized verification sample
42.5%
40/40 responses
IFEVALserialized verification sample
67.5%
40/40 responses
MMLU_PROserialized verification sample
47.5%
40/40 responses

Industry benchmarks

Accuracyattempts · responded
GPQA_DIAMOND GPQA Diamond
41.0%
200 attempts · 190 responded
HUMANEVAL_PLUS HumanEval+
82.5%
40 attempts · 40 responded
IFEVAL IFEval
70.6%
180 attempts · 173 responded
MMLU_PRO MMLU-Pro
38.0%
200 attempts · 172 responded
Industry benchmark Bars = accuracy %, black cap = that score

Custom 9-task suite

Mean /5n=trials
CUSTOM json_strict
5.00
9 trials
CUSTOM reasoning_multistep
5.00
9 trials
CUSTOM code_gen_long
4.11
9 trials
CUSTOM summarize
4.00
9 trials
CUSTOM agentic_tool_use
3.83
6 trials
CUSTOM refactor_existing_code
3.83
6 trials
CUSTOM agentic_prompt
3.33
9 trials
CUSTOM code_debug -
2.79
14 trials
CUSTOM creative_write -
0.00
9 trials
Custom suite · scored 0–5 by consensus (m3 + Fable-5) ★ = task scored a 5 at least once · sorted by mean score

Cross-judge validation

m3 vs Fable-5 Δ

No judge bias detected on this model — m3 and Fable-5 agree across all tasks.

Per-task distribution

Score

The accumulated custom suite below uses minimax-m3 + claude-fable-5 consensus. The dedicated 2026-08-20 campaign above is supplemental and uses minimax-m3 + a blind grok-4.6 re-judge on every successful response. Industry suites use deterministic family-specific graders. See methodology and the campaign report.