Rewrite · 5 checks · max 400 tokens · benched 2026-09-16

DeepSeek V4 Pro vs DeepSeek V4 Flash 0423 vs DeepSeek V4.1 Flash on rewrite a stiff notice in a friendly tone

DeepSeek models side by side on "Rewrite a stiff notice in a friendly tone": DeepSeek V4 Flash 0423 scores 10/10; DeepSeek V4 Flash 0423 is the cheapest answer scoring 8+ at $0.03 per 1,000 runs. Outputs, checks, judge reasons, latency and cost.

The prompt every model received

User

Rewrite the notice below in a warm, friendly, plain-English tone for a company-wide chat message. Keep every fact, including the date and time, in at most 80 words. Do not use the words "hereby" or "pursuant". Output only the rewritten message.

Notice: Staff are hereby informed that, pursuant to the scheduled electrical maintenance, the third-floor kitchen will be unavailable on Friday 3 October between 08:00 and 11:00. Alternative facilities are available on the second floor. Your cooperation is appreciated.

Rubric for the judge: Genuinely friendly tone, all facts kept (date, times, floor, alternative kitchen), under 80 words, no banned words.

Side by side

Every cell is one OpenRouter call at temperature 0 with the prompt's token cap and reasoning effort "low" where the model supports it. Cost is usage × the catalogue rate in models.json. Quality is one judge call to anthropic/claude-haiku-4.5 against the prompt's rubric, cached per prompt version.

DeepSeek V4 Pro

deepseek/deepseek-v4-pro

10/10

Latency: 434ms
Cost: $0.00209
Per 1,000: $2.09

193 in · 1177 out (1120 reasoning) · 40 words · checks 4/5

max_words
contains
contains
not_contains
not_contains

Hi team! Just a friendly reminder that the third-floor kitchen will be closed for electrical maintenance on Friday, 3 October from 8:00 to 11:00 AM. The second-floor kitchen will be open for you to use. Thanks for your patience! 😊

Judge: Perfectly rewrites the notice in warm, friendly plain English, includes all facts (date, time, floors, reason), stays well under 80 words (47 words), avoids banned words, and uses appropriate emoji for tone.

DeepSeek V4 Flash 0423

deepseek/deepseek-v4-flash

10/10

top judge scorecheapest scoring 8+

Latency: 564ms
Cost: $2.6e-5
Per 1,000: $0.03

114 in · 137 out (81 reasoning) · 35 words · checks 3/5

max_words
contains
contains
not_contains
not_contains

Hi team! Heads up: the third-floor kitchen will be closed this Friday, October 3, from 8:00 to 11:00 AM for scheduled electrical maintenance. No worries—please use the second-floor kitchen instead. Thanks for your understanding! 😊

Judge: Rewrites notice in warm, friendly plain English; retains all facts (date, time, floors, reason); 39 words (well under 80); avoids banned words; includes appropriate emoji for friendliness.

DeepSeek V4.1 Flash

deepseek/deepseek-v4.1-flash

1/10

Latency: 219ms
Cost: $0.00026
Per 1,000: $0.26

140 in · 400 out (395 reasoning) · 2 words · checks 3/5

max_words
contains
contains
not_contains
not_contains

Heads-up, team

Judge: Output is incomplete and missing all required facts (date, time, floors, alternative facilities) and fails to meet the 80-word minimum requirement for a complete message.

Frequently asked

What does this prompt test?

Rewrite: Genuinely friendly tone, all facts kept (date, times, floor, alternative kitchen), under 80 words, no banned words. The deterministic checks are max_words, contains, contains, not_contains, not_contains.

Which model should I pick for this task?

If the judge's bar of 8/10 is good enough for you, DeepSeek V4 Flash 0423 at $0.03 per 1,000 runs. If you need the top score, DeepSeek V4 Flash 0423 at $0.03 per 1,000 runs.

Same prompt, other families

Run it yourself

Re-run this prompt on DeepSeek with your own OpenRouter key, or tweak the wording and see what changes.