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. Claude Opus 5.5 (max) leads overall, 91.8 to 88.7.
Claude Opus 5.5 (max effort) wins this one, and the board is not close: #1 at 2845.8 Elo and 91.77 overall, against #15, 2256.0 Elo and 88.73 overall for GLM-5.3. The 95% Elo intervals do not overlap (2790.8 to 2899.6 against 2220.4 to 2296.8). Where Claude Opus 5.5 (max effort) pulls ahead: Length Adherence (97.62 vs 90.43) and Visual Cue Quality (89.31 vs 83.51). GLM-5.3 does not take a single metric off Claude Opus 5.5 (max effort) in this pairing. The trade is price: GLM-5.3 costs about $0.114 per article against $3.433 for Claude Opus 5.5 (max effort), so you pay for the higher rank. GLM-5.3 publishes open weights; Claude Opus 5.5 (max effort) does not. Default to Claude Opus 5.5 (max effort). Reach for GLM-5.3 when the lower price and open weights matter more to you than the headline rank.
Pick Claude Opus 5.5 (max effort) for the stronger board result, length adherence, and visual cues.
Pick GLM-5.3 for the lower price 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.
Blue bars: Claude Opus 5.5 (max). Orange bars: GLM-5.3. Same 0–100 scale; the bold bar wins that metric.
| Claude Opus 5.5 (max) | GLM-5.3 | |
|---|---|---|
| Overall / 100 | 91.8 | 88.7 |
| Writing Elo | 2846 | 2256 |
| Run-to-run spread (± overall std) | 0.850 | 1.180 |
| Cost per script (USD) | 3.433 | 0.114 |
| Avg latency (s) | 1,023.3 | 312.8 |
| Open weights | No | Yes |
Full scorecards: Claude Opus 5.5 (max) · GLM-5.3. How scoring works: methodology.
← All comparisons