Towards AITowards AIToneBench

Qwen3.8 Flash vs MiniMax M3

Head-to-head on the Towards AI writing benchmark: same 10 real YouTube scripts, five runs each, scored blind by a three-family judge panel. Qwen3.8 Flash leads overall, 83.0 to 82.7. Their per-model Elo intervals overlap on this board.

Qwen3.8 Flash
#65 Elo 1699 · 83.0/100
MiniMax M3
#71 Elo 1672 · 82.7/100
Cost / script
$0.013 vs $0.023
Human baseline
90.9 Qwen3.8 Flash falls below it · MiniMax M3 falls below it

The verdict

This one is closer than the ranks suggest. Qwen3.8 Flash sits at #65 with 1698.8 Elo and 82.97 overall; MiniMax M3 is at #71 with 1672.0 Elo and 82.70 overall. Their 95% Elo intervals overlap (1650.3 to 1754.4 against 1632.1 to 1702.1), so do not read much into the exact order. Where Qwen3.8 Flash pulls ahead: Slop Score (EQ-Bench + ours) (88.61 vs 85.83) and Length Adherence (92.46 vs 89.72). MiniMax M3 still wins on Visual Cue Quality (82.42 vs 79.55) and Writing Craft & Clarity (83.89 vs 82.19), so it is not a clean sweep. The trade is price: Qwen3.8 Flash costs about $0.013 per article against $0.023 for MiniMax M3, so you pay less for the higher rank. MiniMax M3 publishes open weights; Qwen3.8 Flash does not. Default to Qwen3.8 Flash. Reach for MiniMax M3 when visual cues, writing craft, and open weights matter more to you than the headline rank.

Pick Qwen3.8 Flash for the stronger board result, slop score, and length adherence.
Pick MiniMax M3 for visual cues, writing craft, and open weights.

Reading the Elo intervals. Each range estimates uncertainty around one model's Elo. We repeatedly sample from every model's recorded run scores on each task and recalculate the full ranking. Comparing these separate ranges does not test the Elo gap between two models. The tasks stay fixed, so the ranges do not measure how the ranking would change on new tasks.

Metric by metric

Blue bars: Qwen3.8 Flash. Orange bars: MiniMax M3. Same 0–100 scale; the bold bar wins that metric.

Tone & Voice Match19% weight
81.9
82.2
Writing Craft & Clarity13% weight
82.2
83.9
Substance, Accuracy & Value15% weight
83.2
81.4
Continuity & Emotion14% weight
78.7
79.6
YouTube Best Practices12% weight
82.2
79.8
Hook Strength10% weight
83.5
84.8
Length Adherence8% weight
92.5
89.7
Slop Score (EQ-Bench + ours)5% weight
88.6
85.8
Visual Cue Quality4% weight
79.5
82.4

Everything else that differs

Qwen3.8 FlashMiniMax M3
Overall / 10083.082.7
Writing Elo16991672
Run-to-run spread (± overall std)2.4201.980
Cost per script (USD)0.0130.023
Avg latency (s)341.7109.9
Open weightsNoYes

Full scorecards: Qwen3.8 Flash · MiniMax M3. How scoring works: methodology.

← All comparisons