ds DeepSeek V3.2
DeepSeek · open weights · writing benchmark
The short version
DeepSeek V3.2 sits at #77 of 130 on ToneBench, toward the value end of the board, with a writing Elo of 1410 and an overall score of 77.2 out of 100. We measured it by having it write all 9 of our real YouTube scripts, five times each, then scoring every draft blind against our own finished versions. Here is how it shook out.
Within DeepSeek it ranks 7th of 8. DeepSeek V4 Flash 0731 is the family's top writer here, about 545 Elo ahead, so this config trades some quality for whatever it saves you in cost or speed.
Among open-weights models, it comes in 18th of 47. Open weights still trail the closed frontier on pure voice fidelity, and you can see that in the gap at the top.
What it does best is Writing Craft: it ranks 67th on that dimension at 81.0, above the board average. Its softer spot is Anti-Slop (77.2, 98th), which is the thing to watch if that metric matters most for your use.
It was uneven across the 9 scripts. Its best run was the founder announcement / personal origin story (84.0) and its weakest was the personal technical walkthrough / agentic workflow case study (68.5), a spread of about 16 points. Worth knowing if your writing skews toward one of those formats.
On cost, it runs about $0.004/task. DeepSeek V4 Flash (native chat alias) scores higher for less money, at $0.002/task, so on pure value this config is not the frontier. Still, it beats 29 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: Flow & Emotion (+2.9); furthest behind: Anti-Slop (-5.5).
Per-metric scores
Nine writing dimensions, each scored 0–100 and blended by the editorial weight shown. Rank is against all 130 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 9 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, DeepSeek V3.2 varies by ± 4.7 points between runs versus a board median of ± 3.8, so it is about as repeatable as the typical model on the board.
Its steadiest is Length Discipline (± 5.1 vs ± 8.6 board median), which you can rely on run after run.
| Metric | This model | Board median | Verdict |
|---|---|---|---|
| Tone & Voice ? | 77.9 ± 2.6 | ± 2.2 | typical |
| Writing Craft ? | 81.0 ± 2.3 | ± 2.0 | typical |
| Substance & Value ? | 80.3 ± 2.7 | ± 2.6 | typical |
| Flow & Emotion ? | 77.3 ± 2.7 | ± 2.6 | typical |
| YouTube Structure ? | 74.0 ± 4.1 | ± 3.6 | typical |
| Hook ? | 81.0 ± 4.0 | ± 3.0 | typical |
| Length Discipline ? | 65.3 ± 5.1 | ± 8.6 | steadier than most |
| Anti-Slop ? | 77.2 ± 3.5 | ± 2.6 | typical |
| Visual Cues ? | 74.1 ± 6.1 | ± 4.7 | 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 77.2 is the consensus of three family-disjoint judges scoring the same 45 stored drafts blind with the identical rubric. The highest and lowest judge differ by 8.2 points on its overall. A typical amount of judge disagreement for this board. How the panel works: methodology.
How we ran DeepSeek V3.2
- Accessed viaOpenRouter · model page ↗
- Exact model / route id
deepseek/deepseek-v3.2-exp(provider default reasoning) - Audited route overrideOpenRouter ·
deepseek/deepseek-v3.2-expforautomated_video_workflow(Task-7 exact OpenRouter endpoint pin novita/fp8) - Generations5 independent runs × 9 scripts = 45 drafts, stored immutably after generation
- Run dates2026-07-29 → 2026-08-11
- 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 DeepSeek V3.2 at writing?
On ToneBench it ranks #77 of 130 with a writing Elo of 1410 and an overall score of 77.2/100. That score comes from writing our 9 real YouTube scripts five times each and scoring every draft blind against our own finished versions across nine writing dimensions.
Is DeepSeek V3.2 the best DeepSeek model for writing?
Not quite. Within DeepSeek it ranks 7th of 8; DeepSeek V4 Flash 0731 is the family's best writer here.
Is DeepSeek V3.2 good value for the money?
It costs about $0.004/task. DeepSeek V4 Flash (native chat alias) scores higher for less, so it is not the value pick.
What are DeepSeek V3.2's strengths and weaknesses?
Its strongest dimension is Writing Craft (67th on the board, 81.0). Its weakest is Anti-Slop (98th on the board, 77.2). The full nine-metric breakdown is on this page.
Does DeepSeek V3.2 write some formats better than others?
Yes. Its best of our 9 scripts was the founder announcement / personal origin story (84.0) and its weakest was the personal technical walkthrough / agentic workflow case study (68.5).
How consistent is DeepSeek V3.2 between runs?
We run every script 5 times. DeepSeek V3.2's overall score varies by about ±4.7 points between runs, versus a board median of ±3.8. That is typical consistency for this board. The full per-metric variability table is on this page.
How was DeepSeek V3.2 evaluated?
Via OpenRouter using the exact model/route id deepseek/deepseek-v3.2-exp, run on 2026-07-29. 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 DeepSeek V3.2 open source?
Yes, it is an open-weights model. Among open-weights models it ranks 18th of 47 for writing.