skynet · reasoning
SKYNET

minimax-m2.7-highspeed

Skynet speed-prior m2.7

⚖ COMPARE MODELS
VALS INDEX
59.8% +/- 17.8
Accuracy
VALS INDEX
26.1s
Latency (median)
$VALS INDEX
$0.0621
Cost / Test (1P-est)
SCORE STATUSscored
TEST SETS13
RECORDED TRIALS630
CUSTOM PASS66%
DEEP VALIDATION · 2026-08-20

Dual-judge family campaign

FULL REPORT →
CONSENSUS MEAN4.011 / 5
BOTH-JUDGE PASS71.1%
CROSS-JUDGED45 / 45
MEDIAN WALL25.61s
GPQA_DIAMONDserialized verification sample
48.0%
25/25 responses
IFEVALserialized verification sample
72.0%
25/25 responses
MMLU_PROserialized verification sample
60.0%
25/25 responses

Industry benchmarks

Accuracyattempts · responded
GPQA_DIAMOND GPQA Diamond
45.6%
180 attempts · 180 responded
HUMANEVAL_PLUS HumanEval+
80.0%
40 attempts · 40 responded
IFEVAL IFEval
69.4%
130 attempts · 130 responded
MMLU_PRO MMLU-Pro
44.0%
200 attempts · 200 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
4.56
9 trials
CUSTOM refactor_existing_code
4.33
6 trials
CUSTOM summarize
4.33
9 trials
CUSTOM agentic_tool_use
3.67
6 trials
CUSTOM agentic_prompt
2.78
9 trials
CUSTOM code_debug
2.43
14 trials
CUSTOM creative_write -
0.56
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.