DeepSeek V4.1 Flash · Writing

DeepSeek V4.1 Flash for writing: our test runs

DeepSeek V4.1 Flash ran our 5 writing prompts in llmwise, through the same pipeline your messages take. Here's every reply as it came, whether it passed, what it cost and what it counts as on Pro, next to the other models.

Start free See pricing

Based on 5 of our test runs on October 8, 2026, through OpenRouter with the app's own prompt and settings. Updated October 15, 2026.

Short answer

In our test runs on October 8, 2026, DeepSeek V4.1 Flash passed 4 of 5 of our writing prompts (both hard ones): tied 4th of 20 models. Each reply counted as 1 message on Pro, where it has 60 a day, and cost $0.0005 to run on average.

Prompt by prompt

Each of the 5 writing prompts, and how DeepSeek V4.1 Flash's reply did.

DeepSeek V4.1 Flash on each writing prompt
Prompt	Result	Cost	Time
Announce a second bakery shop on LinkedInEveryday	Passed: Graded 4.8 of 5 on average (lowest 4).	$0.0004	1.3 s
Rewrite corporate jargon in plain wordsEveryday	Failed: Graded 3.7 of 5 on average (lowest 3).	$0.0003	0.9 s
Decline a meeting and offer two timesEveryday	Passed: Graded 4.7 of 5 on average (lowest 4).	$0.0002	2.2 s
A product announcement with five rulesHard	Passed: Graded 4.3 of 5 on average (lowest 4).	$0.0003	0.8 s
Argue both sides of free busesHard	Passed: Graded 4.3 of 5 on average (lowest 4).	$0.0011	2.6 s

Every writing reply from DeepSeek V4.1 Flash

Announce a second bakery shop on LinkedIn

An everyday prompt: “Announce a second bakery shop on LinkedIn”, in full.

Passed: Graded 4.8 of 5 on average (lowest 4).

Some news we've been sitting on for a while, and it feels good to finally say it out loud: Crumb & Co. is opening a second shop in Leeds on 14 November.

We started with one small counter, a handful of recipes, and a lot of help from people who chose to spend their mornings with us. That's the only reason this is happening.

The new shop will have the same bread, the same cakes, and the same faces you know, plus a few of our team who've been waiting for their own chance to run a counter.

To say thank you, the first coffee is on us on opening day. No catch, no vouchers, just come in and have one.

We'd love to see you there. Bring a friend, or don't. Either way, there'll be something warm waiting.

— Crumb & Co., Leeds, 14 November

428 tokens in, 202 out (19 of them reasoning) · 1.3 s · $0.0004 · 1 message on Pro · answered by deepseek/deepseek-v4.1-flash via Together · October 7, 2026

Rewrite corporate jargon in plain words

An everyday prompt: “Rewrite corporate jargon in plain words”, in full.

Failed: Graded 3.7 of 5 on average (lowest 3).

In the third quarter, teams from across the company worked together on a project to make the customer the main focus. The results: retention measures rose 12%, and the rate at which customers left fell by 30 basis points (0.30 percentage points) in the enterprise segment. We also adjusted the size of our channels — online, in-store, phone, and others — so we could use our resources more efficiently.

459 tokens in, 195 out (108 of them reasoning) · 0.9 s · $0.0003 · 1 message on Pro · answered by deepseek/deepseek-v4.1-flash via Together · October 7, 2026

Decline a meeting and offer two times

An everyday prompt: “Decline a meeting and offer two times”, in full.

Passed: Graded 4.7 of 5 on average (lowest 4).

**Subject:** Re: Vendor demo invitation — Wednesday

Hi Tom,

Thanks for the invitation to the vendor demo on Wednesday at 11:00. Unfortunately I can't make it — I'm presenting to the board at that time.

I'd still like to attend, if the demo can be rescheduled. Would Thursday at 10:00 or Friday at 14:00 work? Either suits me.

If neither fits the vendor's schedule, let me know some other options and I'll do my best to work around them.

Apologies for the inconvenience, and thanks for thinking of me.

Best regards,
[Your name]

434 tokens in, 135 out (6 of them reasoning) · 2.2 s · $0.0002 · 1 message on Pro · answered by deepseek/deepseek-v4.1-flash via Together · October 8, 2026

A product announcement with five rules

A hard prompt: “A product announcement with five rules”, in full.

Passed: Graded 4.3 of 5 on average (lowest 4).

**Introducing the Nordvik 2 desk lamp**

The Nordvik 2 is built for desks that work as hard as you do. Its arm turns a full 270 degrees, so light lands exactly where you need it, and five dimming steps take you from focused task lighting to a softer evening glow. A USB-C port keeps it charged and cable clutter to a minimum. It costs €79 and ships on 1 December.

Would you like to see it in person before then?

446 tokens in, 188 out (87 of them reasoning) · 0.8 s · $0.0003 · 1 message on Pro · answered by deepseek/deepseek-v4.1-flash via Together · October 8, 2026

Argue both sides of free buses

A hard prompt: “Argue both sides of free buses”, in full.

Passed: Graded 4.3 of 5 on average (lowest 4).

Free buses make a city fairer and faster. Fares cost more to collect than they raise, so abolishing them saves money while filling seats. Riders who now drive will switch, cutting congestion, emissions, and parking demand. Low-income workers gain access to jobs; students and elderly riders gain independence. Downtown shops gain customers. A city that moves everyone freely is safer, cleaner, and more equal. The bus becomes public infrastructure, like a sidewalk — not a toll road.

Free buses are not free; someone pays. Subsidies come from taxes, pulling money from schools, housing, or health. Packed buses with no revenue still need more vehicles and drivers, so service quality falls. Riders who can afford fares — most of them — get a subsidy they don't need, while the poorest may see worse, slower service. Free transit also mostly serves existing riders rather than luring drivers out of cars. Better to fund frequent, reliable buses and discount fares for those who need help.

423 tokens in, 859 out (655 of them reasoning) · 2.6 s · $0.0011 · 1 message on Pro · answered by deepseek/deepseek-v4.1-flash via Together · October 8, 2026

How it compares on writing

DeepSeek V4.1 Flash beside the job's picks among every model. 3 models share the hard-problems pick, and DeepSeek V4.1 Flash isn't one of them.

DeepSeek V4.1 Flash beside other models on writing
Model	Passed	On Pro	Cost per reply
DeepSeek V4 ProOur pick: hard problems (shared), best value	5 of 5	250 a month on Pro	$0.0011
GPT-6 LunaOur pick: hard problems (shared), everyday	5 of 5	60 a day on Pro	$0.0001
DeepSeek V4.1 Flash	4 of 5	60 a day on Pro	$0.0005

On Pro: Pro's count on each model. Cost: what OpenRouter charged us per reply, on average.

Every limit is published. Paid plans also have a monthly fair-use limit on AI cost: Plus $3.75, Pro $7.50, Max $20, Ultra $42, Studio $85. Using every message on your plan at typical sizes stays under it; very large messages and heavy research use it faster. Every limit, explained.

DeepSeek V4.1 Flash next to each model on writing

DeepSeek V4.1 Flash against each model on the same 5 writing prompts. One run each, through OpenRouter: a wait depends on the provider and the load that day, so a lead under 10% counts as close.

Against Claude Fable 5.1: DeepSeek V4.1 Flash answered 4.5× sooner (1.3 s to 6.0 s), cost 38.3× less, and passed 4 of 5 to its 3. Only DeepSeek V4.1 Flash passed A product announcement with five rules; Argue both sides of free buses. Only Claude Fable 5.1 passed Rewrite corporate jargon in plain words.
Against Claude Opus 5.5: DeepSeek V4.1 Flash answered 6.2× sooner (1.3 s to 8.2 s), cost 41.0× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed Argue both sides of free buses. Only Claude Opus 5.5 passed Rewrite corporate jargon in plain words.
Against Claude Sonnet 5.5: DeepSeek V4.1 Flash answered 1.8× sooner (1.3 s to 2.3 s), cost 7.6× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed Argue both sides of free buses. Only Claude Sonnet 5.5 passed Rewrite corporate jargon in plain words.
Against Claude Sonnet 5: DeepSeek V4.1 Flash answered 3.3× sooner (1.3 s to 4.5 s), cost 7.6× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed Announce a second bakery shop on LinkedIn. Only Claude Sonnet 5 passed Rewrite corporate jargon in plain words.
Against Claude Haiku 5.5: DeepSeek V4.1 Flash answered 1.1× sooner (1.3 s to 1.5 s), cost 2.8× more, and passed 4 of 5 to its 3. Only DeepSeek V4.1 Flash passed Argue both sides of free buses.
Against Claude Haiku 4.5: DeepSeek V4.1 Flash answered 1.7× sooner (1.3 s to 2.2 s), cost 2.5× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed A product announcement with five rules. Only Claude Haiku 4.5 passed Rewrite corporate jargon in plain words.
Against GPT-6 Astra: DeepSeek V4.1 Flash answered 3.6× sooner (1.3 s to 4.8 s), cost 24.7× less, and passed 4 of 5 to its 5. Only GPT-6 Astra passed Rewrite corporate jargon in plain words.
Against GPT-6.1 Sol: DeepSeek V4.1 Flash answered 2.4× sooner (1.3 s to 3.3 s), cost 2.5× less, and passed 4 of 5 to its 4.
Against GPT-6 Sol: DeepSeek V4.1 Flash answered 2.6× sooner (1.3 s to 3.5 s), cost 5.0× less, and passed 4 of 5 to its 4.
Against GPT-6 Luna: DeepSeek V4.1 Flash answered 1.8× sooner (1.3 s to 2.4 s), cost 4.5× more, and passed 4 of 5 to its 5. Only GPT-6 Luna passed Rewrite corporate jargon in plain words.
Against Gemini 4 Argon: DeepSeek V4.1 Flash answered 4.3× sooner (1.3 s to 5.7 s), cost 3.1× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed Argue both sides of free buses. Only Gemini 4 Argon passed Rewrite corporate jargon in plain words.
Against Gemini 3.1 Pro (preview): DeepSeek V4.1 Flash answered 6.8× sooner (1.3 s to 9.1 s), cost 20.4× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed Argue both sides of free buses. Only Gemini 3.1 Pro (preview) passed Rewrite corporate jargon in plain words.
Against Gemini 3.8 Flash: DeepSeek V4.1 Flash answered 2.9× sooner (1.3 s to 3.9 s), cost 2.9× less, and passed 4 of 5 to its 3. Only DeepSeek V4.1 Flash passed A product announcement with five rules.
Against DeepSeek V4 Pro: DeepSeek V4.1 Flash answered 2.9× sooner (1.3 s to 3.9 s), cost 2.4× less, and passed 4 of 5 to its 5. Only DeepSeek V4 Pro passed Rewrite corporate jargon in plain words.
Against Grok 4.7: DeepSeek V4.1 Flash answered 4.5× sooner (1.3 s to 6.0 s), cost 12.8× less, and passed 4 of 5 to its 3. Only DeepSeek V4.1 Flash passed Argue both sides of free buses.
Against Kimi K3: DeepSeek V4.1 Flash answered 2.2× sooner (1.3 s to 3.0 s), cost 8.5× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed A product announcement with five rules. Only Kimi K3 passed Rewrite corporate jargon in plain words.
Against GLM 5.3: DeepSeek V4.1 Flash answered 1.3× sooner (1.3 s to 1.8 s), cost 1.3× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed Argue both sides of free buses. Only GLM 5.3 passed Rewrite corporate jargon in plain words.
Against GLM 5.3 Flash: DeepSeek V4.1 Flash answered 6.3× sooner (1.3 s to 8.5 s), cost 1.4× more, and passed 4 of 5 to its 2. Only DeepSeek V4.1 Flash passed A product announcement with five rules; Argue both sides of free buses.
Against Mistral Large 4: DeepSeek V4.1 Flash answered 7.4× sooner (1.3 s to 9.8 s), cost 5.3× less, and passed 4 of 5 to its 4. Only DeepSeek V4.1 Flash passed A product announcement with five rules. Only Mistral Large 4 passed Rewrite corporate jargon in plain words.

How these runs were done

Rubric (graded). The grader model scores the reply from 1 to 5 on each published criterion. It passes with an average of 4 or more and no criterion under 3, and only if it also meets the prompt's automatic rules (length, words it must or mustn't use).

The grader is Claude Opus 5.5 at low reasoning effort; its own replies are graded by GPT-6 Astra, so no model grades itself. Its prompt and every rubric are on the methods page.

How the runs were done, and every writing prompt.

DeepSeek V4.1 Flash, and writing, elsewhere

Questions

Is DeepSeek V4.1 Flash good for writing?

In our test runs it passed 4 of 5 writing prompts, tied 4th of the 20 models in llmwise. Every reply is on this page, so you can judge them yourself.

How many of my messages does a writing reply from DeepSeek V4.1 Flash use?

1 message each on Pro, where it has 60 a day on Pro. The price of a message is fixed and shown before you send it, however long the reply.

How were these runs done?

The same way for every model: each prompt sent through llmwise's own pipeline, each reply checked the same way. The methods page has every prompt and how each is scored.

Claude, GPT, Gemini, DeepSeek, Grok, Kimi, GLM, and Mistral, in one chat.

See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.

Start free See pricing