Model Comparison
GLM-5.3 (max) leads DeepSeek V4.1 Flash (max) by 177 Elo, and their 95% ranges sit 98 Elo apart. Among 84 models, each ranked by its best configuration, GLM-5.3 is #10 and DeepSeek V4.1 Flash is #16. Cost per script: $0.240 for GLM-5.3 (max) and $0.017 for DeepSeek V4.1 Flash (max).
Last updated , from board data generated at 20:15 UTC.
PickGLM-5.3
GLM-5.3 (max)
Pick GLM-5.3 (max) for the scripts that go on camera, since its range sits clear of every DeepSeek setting at $0.240 a script.
PickDeepSeek V4.1 Flash
DeepSeek V4.1 Flash (high)
Pick DeepSeek V4.1 Flash (high) for quick first drafts at $0.012 and about a minute each, since its range overlaps max's.
DeepSeek V4.1 Flash is the cheaper writer by a wide margin, and GLM-5.3 is the better one. Ranked by best setting, GLM-5.3 is the #10 model of 84 and DeepSeek V4.1 Flash the #16 model. GLM-5.3 (max) sits 177 Elo above DeepSeek V4.1 Flash (max), and their 95% ranges are separate, which is strong evidence for that order. GLM wins all ten scripts against every DeepSeek setting, with the clearest gaps on hooks and visual cues, more than three points each at max. That lead costs about 14 times as much, $0.240 a script against $0.017, and 6.1 minutes against 2.4. On DeepSeek's side, the thinking levels matter a lot. Going from no thinking to max adds 397 Elo while staying under two cents a script, so I'd never switch it off. One step down, DeepSeek V4.1 Flash (high) costs $0.012, finishes in about a minute, and its range overlaps max. At max, DeepSeek also lands closer to the requested length and its drafts are a touch steadier. My split: GLM-5.3 for the final version, DeepSeek at high for everything before it.
Disclosure: GLM-5.3 comes from the same lab as one of the four judges, GLM 5.3 Flash. The other three sit outside its lab, so it is never scored by its own lab alone (how the panel works). The judge analysis measures how much each judge favors its own lab.
Writing Score and Writing Elo. The Writing Score (0 to 100) is the weighted average of the nine metric scores: seven from the judges, Length Adherence from code and the Slop Score half from each. The Writing Elo is a rating from pairwise comparisons: on each script, each configuration's average across its drafts is compared with every other configuration's, and gaps within the draft-to-draft variation count as draws. Elo rewards winning often, so the two can put configurations in a different order.
Elo ranges. Each 95% range shows how far a configuration's Elo could move if the drafts had come out slightly differently. It comes from resampling every configuration's recorded draft scores on each reference script and ranking again. Comparing two separate ranges does not test the Elo gap between them: ranges far apart are strong evidence for the order, and overlapping ranges do not mean a tie. The reference scripts stay fixed, so the ranges say nothing about new scripts. The methodology page has the full description.
Every ranked configuration of GLM-5.3 and DeepSeek V4.1 Flash, scored on the same 10 scripts by the same four judges.
Shaded rows are the configurations chosen for the metric comparison below.
| Configuration | Writing Elo | 95% Range | Writing Score | Cost / Script | Time / Script | Reasoning Tokens | Configuration Rank |
|---|---|---|---|---|---|---|---|
| GLM-5.3 (max) | 2178 | 2138 to 2211 | 82.4 | $0.240 | 6.1 min | 46,444 | #18 |
| DeepSeek V4.1 Flash (max) | 2001 | 1957 to 2040 | 81.0 | $0.017 | 2.4 min | 22,010 | #31 |
| DeepSeek V4.1 Flash (high) | 1937 | 1898 to 1976 | 80.3 | $0.012 | 1.1 min | 13,809 | #41 |
| DeepSeek V4.1 Flash (low) | 1832 | 1782 to 1880 | 79.3 | $0.0077 | 0.7 min | 6,740 | #53 |
| DeepSeek V4.1 Flash (no thinking) | 1604 | 1561 to 1647 | 76.8 | $0.0037 | 0.3 min | 0 | #72 |
Cost is one script at the API list price of the route used, uncached and without judging. Time is the average for one accepted script, retries included.
Each finding is computed from the configurations above. The 95% ranges describe individual configurations; overlap alone establishes neither a tie nor an order for a pair.
Pick one configuration of each model to compare their metric scores, consistency across drafts and per-script scores. Each menu starts on the model's best configuration.
GLM-5.3 (max) scores higher on eight of the nine metrics and DeepSeek V4.1 Flash (max) on one. Weighted by each metric's share of the Writing Score, Hook Strength does the most to put GLM-5.3 (max) ahead, with a 3.3-point lead on that metric. The weight beside each metric is its share of the Writing Score.
| GLM-5.3 (max) | DeepSeek V4.1 Flash (max) | |
|---|---|---|
| Score spread across drafts (±) | 1.39 | 1.30 |
| Lowest script score | 80.5 Script 10 | 79.7 Script 10 |
| Highest script score | 84.1 Script 7 | 82.4 Script 7 |
| Range across scripts | 3.6 | 2.7 |
Spread is how much the Writing Score moves between a configuration's drafts, and lower is steadier. Green marks the better number in each row.
Writing Score on each reference script. Gap is GLM-5.3 (max) minus DeepSeek V4.1 Flash (max), and the higher score is in bold.
| Script | GLM-5.3 (max) | DeepSeek V4.1 Flash (max) | Gap |
|---|---|---|---|
| Script 1 opinion / warning explainer | 82.1 | 80.8 | +1.3 |
| Script 2 news-analysis / skeptical explainer | 82.5 | 81.2 | +1.3 |
| Script 3 personal roadmap / opinion | 81.5 | 81.2 | +0.3 |
| Script 4 short explainer | 81.0 | 80.1 | +0.9 |
| Script 5 news-analysis / opinion explainer | 82.5 | 81.1 | +1.4 |
| Script 6 founder announcement / personal origin story | 82.9 | 80.8 | +2.1 |
| Script 7 personal technical walkthrough / agentic workflow case study | 84.1 | 82.4 | +1.7 |
| Script 8 product release announcement / personal observation | 82.7 | 80.7 | +2.0 |
| Script 9 career guide / hiring analysis | 84.0 | 82.2 | +1.8 |
| Script 10 engineering process walkthrough / presentation adaptation | 80.5 | 79.7 | +0.8 |
Each configuration name in the table above opens its own scorecard.
Pick a second model. A pair with its own written match-up opens that page, and any other pair opens in the comparison tool.
Every configuration wrote five drafts for each of the same 10 briefs, from the same prompt. Generation was closed-book, with the prompt as the only source and no web search or other tools. Each judge scores every draft once: Claude Opus 5.5 at medium effort through the Claude Code subscription, GPT-6.1 Sol at medium reasoning through the Codex CLI, GLM 5.3 Flash through OpenRouter, and Jev 1.13, a typed judge from TypeSafe that answers each metric as a multiple-choice question. The channel's finished script is their reference for the voice, and it gets no score on the board itself.
← All comparisons