DeepSeek V4.1 Flash (high)
DeepSeek · open weights · writing benchmark
The short version
DeepSeek V4.1 Flash (high) lands in the upper third of ToneBench at #45 of 181, with a writing Elo of 2066 and an overall score of 86.8 out of 100. Those numbers come from it writing its own version of the scripts behind 10 of my What's AI videos, five times each, with every draft scored blind against mine (how the scoring works).
Its Elo comes from comparing its average on each script with every other model's average on the same script, with gaps inside the run-to-run noise counted as draws. The 95% range is 2021 to 2109. That range overlaps 19 other configurations, ranked #31 to #50, so its exact spot inside that band is too close to call on this board. What the ranges can and can't tell you. One disclosure: the DeepSeek V4.1 Flash line also holds one of the four judge seats. The other three judges come from other model families, so no family scores itself alone (how the panel works).
Its biggest edge is Length: 87.6, 47th on the board and about 16 points above the board average. Even its weakest, Cues (79.9), sits above the board average, so there's no real weak spot to plan around.
The format changes the result. Its best score was on the career guide / hiring analysis (89.2) and its weakest on the engineering process walkthrough / presentation adaptation (84.7), about 4 points apart. If most of your scripts look like one of those, weigh that score over the average.
It costs about $0.009 per script. MiMo-V2.6-Flash scores higher and costs about $0.004 per script, so on value this isn't the pick. Still, plenty of pricier configs score lower than it does.
If you're on DeepSeek anyway, DeepSeek V4.1 Flash (max) is worth paying under a cent more per script.
Among open-weights models, it's 8th of 56. The best open-weights config is GLM-5.3 at #12, and everything above it is closed, so open weights still trail the frontier on this kind of writing.
DeepSeek V4.1 Flash is also on the board at other effort settings, and DeepSeek V4.1 Flash (max) writes better, at Elo 2165. Check the thinking-levels page before you lock this one in: it shows what each setting costs and how long it takes.
Skill profile
Read the shape: the further a corner reaches, the stronger the model is on that metric. The dashed line is the board average and the dotted line is the board's best score on each metric, so a corner outside the dashed line beats the average there. Hover a point for the exact numbers.
Metric by metric
Nine writing metrics, each scored 0–100 and blended into the overall by the editorial weight shown next to it. Rank is out of all 181 current-ranked models. The thick bar is this model; the thin line above it is the field's best on that metric (darker) and the one below is the field's average (lighter), on the same scale. The small ± is how much the score moves from run to run, and the ? next to each name says what the metric rewards.
Script by script
The same 10 real scripts every current-ranked model writes, each scored on its own as the mean of 5 runs. Different formats stress different skills: a model can nail a tight explainer and still stumble on a personal story, so look for the format closest to what you write.
Consistency: how much it moves between runs
A model whose drafts swing from run to run is harder to use than its average suggests. Every score on this page is the mean of 5 runs per script, and the ± is the run-to-run standard deviation. Overall, DeepSeek V4.1 Flash (high) moves by ± 1.8 against a board median of ± 2.5, so it's steadier than the typical model here.
| Metric | This model | Board median | Verdict |
|---|---|---|---|
| Tone ? | 87.0 ± 1.4 | ± 1.5 | typical |
| Craft ? | 88.7 ± 0.9 | ± 1.4 | typical |
| Substance ? | 86.1 ± 1.5 | ± 1.7 | typical |
| Flow ? | 86.2 ± 1.3 | ± 1.6 | typical |
| YouTube ? | 85.7 ± 2.1 | ± 2.5 | typical |
| Hook ? | 87.8 ± 1.6 | ± 1.9 | typical |
| Length ? | 87.6 ± 8.0 | ± 8.2 | typical |
| Slop ? | 89.8 ± 2.0 | ± 1.9 | typical |
| Cues ? | 79.9 ± 3.3 | ± 2.8 | typical |
Cost, time and tokens
Logged on every run. None of it touches the writing score, but it's often what decides whether a model fits your pipeline at all. Time is the average time to get one accepted script, retries included, and cost is priced per script at list price.
What each judge scored it
One judge can have taste of its own, which is why there are four: three LLM judges from three model families, plus Jev, a typed judge from TypeSafe, each scoring the same 50 stored drafts blind against the same rubric. The overall of 86.8 combines their scores with the parts computed in code (length, and half of the slop score). The highest and lowest judge are 3.6 points apart on its overall. That's tighter than the board median of 6.3, so the panel mostly agrees on this one. How the panel works.
How we ran DeepSeek V4.1 Flash (high)
- Accessed viaOpenRouter · model page ↗
- Exact model / route id
deepseek/deepseek-v4.1-flashatreasoning_effort: high - Generations5 independent runs × 10 scripts = 50 drafts, stored immutably after generation
- Run dates2026-09-29
- Judges4-judge panel: Claude Opus 5, high effort (Anthropic) + DeepSeek V4.1 Flash (OpenRouter) + GPT-5.6 Sol, medium reasoning (Codex CLI) + Jev 1.13, typed judge (TypeSafe). All four score every draft independently, blind to which model wrote it. The three LLM judges use the identical rubric and have to quote evidence before they score a metric; Jev answers the same rubric as one question per metric, choosing among five described quality levels, and writes no text. Seven metrics are the plain average of the four judges; Length is computed from the spoken word count, and Slop is half computed (the EQ-Bench slop-score pipeline plus our banned-phrase scan) and half judged. The Elo is cross-checked with a Bradley-Terry refit (methodology)
- Sampling and promptprovider-default sampling, no fine-tuning, no system-prompt changes. Every current-ranked model gets the same prompt, about 11,000 tokens (the count shifts a bit with each model's tokenizer): a style guide built from my voice and storytelling guides, the target length, and the brief with its research packet. What changes from row to row is the model configuration
- What stays privatethe exact rubric and any non-public reference material. Scripts are normally benchmarked before their video goes out, which lowers the odds a model has seen mine, though it doesn't make that impossible (why)
Frequently asked questions
How good is DeepSeek V4.1 Flash (high) at writing?
It's #45 of 181 on ToneBench, with a writing Elo of 2066 and an overall score of 86.8/100. It wrote its own version of the scripts behind 10 of my What's AI videos 5 times each, and four judges scored every draft blind against mine: three LLM judges from three model families, plus Jev, a typed judge from TypeSafe. Length and half of the slop score are computed in code.
Is DeepSeek V4.1 Flash (high) the best DeepSeek model for writing?
No. It's 3rd of 15 DeepSeek configs; DeepSeek V4.1 Flash (max) is about 99 Elo ahead.
Is DeepSeek V4.1 Flash (high) good value for the money?
It costs about $0.009 per script. MiMo-V2.6-Flash scores higher and costs less, so it isn't the value pick.
What are DeepSeek V4.1 Flash (high)'s strengths and weaknesses?
Against the board average, its biggest edge is Length (87.6, +15.6) and even its weakest, Cues (79.9, +2.3), is above average. The nine-metric breakdown on this page shows the board average and board best for each.
How was DeepSeek V4.1 Flash (high) evaluated?
Through OpenRouter, using the exact model/route id deepseek/deepseek-v4.1-flash at reasoning_effort high, first run on 2026-09-29. Each script ran 5 times with provider-default sampling and no fine-tuning. The 'How we ran it' section and the methodology page have the rest.
Is DeepSeek V4.1 Flash (high) open source?
Yes, it's an open-weights model. Among open-weights models it's 8th of 56 for writing.