Z GLM-5.3
Zhipu (GLM) · open weights · writing benchmark
The short version
GLM-5.3 sits at #24 of 138 on ToneBench, in the middle of the pack, with a writing Elo of 2042 and an overall score of 86.8 out of 100. We measured it by having it write all 10 of our real YouTube scripts, five times each, then scoring every draft blind against our own finished versions. Here is how it shook out.
Inside Zhipu (GLM), this is the strongest writer we tested. It edges out the other 6 Zhipu (GLM) configs on the board, so if you're staying in this family for voice work, this is the one to reach for.
Among open-weights models, it comes in 3rd of 50. That is a genuinely strong showing for an open-weights model on a voice-and-tone task, which is usually where the closed frontier still pulls ahead.
Worth knowing before you rely on it: this model is swingy. Its overall score moves ± 6.6 points between runs of the same brief, versus a board median of ± 3.1. A great draft and a mediocre one can come from the identical prompt. The shaded bands on the metric bars below show where that volatility lives.
What it does best is Anti-Slop: it ranks 12th on that dimension at 91.6, well above the board average. Its softer spot is Visual Cues (81.4, 54th), which is the thing to watch if that metric matters most for your use.
It was uneven across the 10 scripts. Its best run was the product release announcement / personal observation (89.5) and its weakest was the news-analysis / opinion explainer (81.1), a spread of about 8 points. Worth knowing if your writing skews toward one of those formats.
On cost, it runs about $0.109/task. GPT-5.6 Terra (ultra) scores higher for less money, at $0.068/task, so on pure value this config is not the frontier. Still, it beats 5 models that cost noticeably more.
Skill profile
The shape is the story: the further a corner reaches, the stronger that dimension. Biggest edge over the board: Length Discipline (+19.2); smallest edge: Substance & Value (+7.1).
Per-metric scores
Nine writing dimensions, each scored 0–100 and blended by the editorial weight shown. Rank is against all 138 current-ranked models. The thick bar is this model; the thin lines above and below are the current field's best model (darker) and average (lighter) on the same scale. The small ± number is its run-to-run variation.
Per-article scores
The same 10 real scripts every current-ranked model writes, scored individually. Different formats stress different skills.
Consistency: run-to-run variability
Every cell on this page is the mean of 5 independent runs per script; the ± numbers are the run-to-run standard deviation. Overall, GLM-5.3 varies by ± 6.6 points between runs versus a board median of ± 3.1, so it is swingier than the typical model here, worth knowing if you need repeatable output.
Its most volatile dimension is Writing Craft (± 3.0 vs a board median of ± 1.9): two runs of the same brief can land visibly different writing craft scores.
| Metric | This model | Board median | Verdict |
|---|---|---|---|
| Tone & Voice ? | 87.9 ± 2.2 | ± 2.0 | typical |
| Writing Craft ? | 87.2 ± 3.0 | ± 1.9 | swingier than most |
| Substance & Value ? | 85.9 ± 4.8 | ± 2.4 | swingier than most |
| Flow & Emotion ? | 85.1 ± 4.5 | ± 2.3 | swingier than most |
| YouTube Structure ? | 85.0 ± 5.7 | ± 3.4 | swingier than most |
| Hook ? | 89.5 ± 1.9 | ± 2.5 | typical |
| Length Discipline ? | 87.5 ± 10.4 | ± 9.4 | typical |
| Anti-Slop ? | 91.6 ± 1.4 | ± 2.3 | typical |
| Visual Cues ? | 81.4 ± 4.8 | ± 4.3 | typical |
Measured, not modeled
Numbers we log on every run and rarely talk about. None of these affect the writing scores; cost and latency are informational.
What each judge scored it
The published overall of 86.8 is the consensus of three family-disjoint judges scoring the same 50 stored drafts blind with the identical rubric. The highest and lowest judge differ by 3.7 points on its overall. The three judges essentially agree on this model. How the panel works: methodology.
How we ran GLM-5.3
- Accessed viaOpenRouter · model page ↗
- Exact model / route id
z-ai/glm-5.3(provider default reasoning) - Generations5 independent runs × 10 scripts = 50 drafts, stored immutably after generation
- Run dates2026-08-27
- Judges3-judge panel: Claude Opus 5, high effort (Anthropic) + DeepSeek V4 Flash (OpenRouter) + GPT-5.6 Sol, medium reasoning (Codex CLI): every draft scored independently by all three with the identical rubric, blind to model identity, evidence quotes required before each metric score; published score = per-metric consensus mean, verified with a Bradley-Terry refit (methodology)
- Samplingprovider-default sampling, no fine-tuning, no system-prompt changes; every current-ranked model gets the identical brief, style contract and research packet
- What stays privatethe exact rubric and any non-public reference material; source-publication timing is disclosed for exceptions (why)
Frequently asked questions
How good is GLM-5.3 at writing?
On ToneBench it ranks #24 of 138 with a writing Elo of 2042 and an overall score of 86.8/100. That score comes from writing our 10 real YouTube scripts five times each and scoring every draft blind against our own finished versions across nine writing dimensions.
Is GLM-5.3 the best Zhipu (GLM) model for writing?
Yes. Among the Zhipu (GLM) configs we tested, GLM-5.3 is the strongest writer on the board.
Is GLM-5.3 good value for the money?
It costs about $0.109/task. GPT-5.6 Terra (ultra) scores higher for less, so it is not the value pick.
What are GLM-5.3's strengths and weaknesses?
Its strongest dimension is Anti-Slop (12th on the board, 91.6). Its weakest is Visual Cues (54th on the board, 81.4). The full nine-metric breakdown is on this page.
Does GLM-5.3 write some formats better than others?
Yes. Its best of our 10 scripts was the product release announcement / personal observation (89.5) and its weakest was the news-analysis / opinion explainer (81.1).
How consistent is GLM-5.3 between runs?
We run every script 5 times. GLM-5.3's overall score varies by about ±6.6 points between runs, versus a board median of ±3.1. That is swingier than typical, so expect more draft-to-draft variation. The full per-metric variability table is on this page.
How was GLM-5.3 evaluated?
Via OpenRouter using the exact model/route id z-ai/glm-5.3, run on 2026-08-27. 5 runs per script, provider-default sampling, no fine-tuning; every draft scored blind with a fixed rubric by a three-family judge panel. Full details in the 'How we ran it' section and the methodology.
Is GLM-5.3 open source?
Yes, it is an open-weights model. Among open-weights models it ranks 3rd of 50 for writing.