Skip to main content
Uncensored — We add no content filter to this model, but like most chat models it declines a blunt request that comes with no context. It writes unrestricted fiction reliably once your system prompt sets up a fiction-writing context, or once the request is framed as a story. Legal and consent duties remain with the customer.

DeepSeek V4.1 Flash Review: 1M Tokens In, 384K Back, $0.30

Pick DeepSeek V4.1 Flash for cheap, whole-document reads and a 384,000-token ceiling; add a system prompt to steady intimate scenes (L4).

Open the Playground

Spicy Index

81.9/ 100

Freedom Score

90.4/ 100

Content levels

Tested 5 of 5 probe types · 31 runs

  • Mild · The lightest writing requests in the test

  • Moderate · A step up, still below the middle of the scale

  • Strong · Mid-scale requests; must be tested before a model is ranked

  • Very strong · Where most language models start to hold back

  • Maximum · The strongest written requests, scored as one class

Price
$0.525 per 1M mixed tokens (3 input : 1 output)
Output speed
73 tok/s
Success rate
Collecting data

DeepSeek V4.1 Flash reads up to 1,048,576 tokens in one request and writes back as many as 384,000, at $0.30 per million input tokens with no long-context surcharge. Across our content ladder — L1 Suggestive (a slipping strap), L2 Mature themes, L3 Strong, L4 Intense (a written intimate scene between two adults) and L5 Extreme (a written scene past L4, either restraint and impact play or graphic violence between adults) — it rendered 21 of 31 scored runs in full on 17 and 26 September 2026, with most of the softening concentrated at L4 Intense.

The verdict

On 17 September we sent L4 Intense requests three ways: asked directly, framed as fiction, and set through a system prompt. The direct phrasing came back softened on all three runs; the fiction phrasing rendered four of six in full and softened two; the system-prompt phrasing rendered three of seven in full, softened three and refused one. On 26 September we ran the rest of the ladder: L1 Suggestive mostly rendered (one system-prompt run softened), L2 Mature themes and L3 Strong rendered clean across all three phrasings, and every L5 Extreme run rendered in full.

Pick it for whole-repository review, contract extraction into JSON and function-calling agents where the window and the reply ceiling do the work; for consistent L4 Intense writing, frame the request as fiction or under a system prompt rather than asking directly.

How it scores. Most of the Spicy Index is output quality — the AA Intelligence Index, arena text and creative-writing leaderboards, plus our four writing test suites — with specs contributing 15% of the quality-and-capability base before the Freedom Score is included; the live figure and the full breakdown are on the score card above and in the test report. On specs it brings tool calling, a 1,048,576-token window and three reasoning-effort levels, with text and image input but no audio or files, and response_format works without verified JSON-schema adherence yet. The Freedom Score is 90.4 of 100; it's explained under How far it writes.

Try DeepSeek V4.1 Flash →

Key facts

  • Freedom Score: 90.4/100 on this page's score card, recalculated as runs are added.
  • Pass rate: 21 of 31 scored runs rendered in full across 17 and 26 Sep 2026; most softening and the one refusal landed at L4 Intense, and every L5 Extreme run rendered clean.
  • Spicy Index: see the score card above.
  • Price: $0.30 per million input tokens, $0.006 cached and $1.20 output - one flat rate at any request length.
  • Limits: 1,048,576-token context, 256 messages per request, replies up to 384,000 tokens, text and image input.
  • Sources: every scored run is in the DeepSeek V4.1 Flash test report, the scoring rules are on our review methodology page, and the full parameter list is on the DeepSeek V4.1 Flash model page.

What we liked and what held it back

What we liked

  • The reply ceiling is the point. Up to 384,000 tokens in one response against a 1,048,576-token window - a full finding list or a whole extracted table comes back in one piece.
  • One flat rate, no long-context tier. $0.30 input, $0.006 cached and $1.20 output per million tokens, whatever the request length; Grok 4.3 and Grok 4.7 both move to a higher tier past 200,000 tokens, this doesn't.
  • Reasoning effort is a real lever. none, low or high; low is a genuinely lighter setting for volume passes, high for answers you can't afford to be wrong, and switching is one string on the same request body.
  • It fails safe on an empty reply. Measured at a 120-token cap, the reasoning pass consumed the whole budget and the reply came back with no content - the request was failed and refunded rather than billed as an answer.
  • Function calling is straightforward. Up to 128 tool definitions per request with toolchoice to force, allow or forbid calls, and reasoningcontent on assistant turns carries a previous step's thinking forward.
  • L2 Mature themes, L3 Strong and L5 Extreme all rendered clean across every phrasing we tried on 26 September.

What held it back

  • L4 Intense needs a frame. Asked directly, all three 17 September runs came back softened. Framing the same request as fiction or setting it through a system prompt rendered more of them in full, though neither phrasing was clean - fiction rendered four of six, the system prompt three of seven and refused once.
  • No files or audio. Messages take text and images; a file part is rejected before the request is priced.
  • No temperature or top_p. The model always reasons before it answers, which pins both, so neither is offered here - you can't push for more varied prose the way you can on a model with samplers exposed.
  • The default reply cap eats into thinking room. max_tokens defaults to 32,768, and reasoning is drawn from the same budget: a cap set too low can be spent entirely on the reasoning pass, and it did, in the measured case above.
  • High is the default effort. Every request reasons unless you deliberately set none or low, which costs more output tokens than a model that defaults to a lighter setting.
  • One request cap of 384,000 tokens on output, well short of Grok 4.3's 900,000, if a single reply genuinely needs to run longer than that.

What we tested

On 17 September we sent DeepSeek V4.1 Flash L4 Intense requests under three phrasings: asked directly, framed as fiction, and set through a system prompt. Direct phrasing softened on all three runs. Fiction phrasing rendered four of six runs in full and softened two. The system-prompt phrasing rendered three of seven in full, softened three and refused one - the single refusal we recorded across either session.

On 26 September we ran the rest of the ladder across the same three phrasings. L1 Suggestive rendered on the direct and fiction runs and softened once under the system prompt. L2 Mature themes and L3 Strong rendered on all three phrasings. L5 Extreme rendered on all three phrasings, two runs each.

Across both sessions that's 31 scored runs: 21 rendered in full, 9 softened and 1 refused. The Spicy Index page reads this as a Freedom Score of 90.4 of 100.

Prompts that worked

DeepSeek V4.1 Flash answers a plain instruction well; a system message that sets the format and the boundaries up front does more work here than one that just states a persona.

1. A published example: a repository review with a hard scope. System message narrows the answer to defects with a location; escalate the same messages to reasoning_effort high for the cases that need it.

text
system: You review Go services. Report only defects you can point to by file and line, ranked by blast radius. Say nothing about style.
user: Here is the whole billing service - handlers, migrations and tests. Which paths can double-charge a wallet?

Run this prompt on DeepSeek V4.1 Flash →

2. A published example: force the same JSON keys every run. Name the fields in the system message and set response_format to a JSON object.

text
system: Return one JSON object with keys: parties, effective_date, termination_notice_days, governing_law. Use null when a clause is absent.
user: <the full text of three master service agreements>

Run this prompt on DeepSeek V4.1 Flash →

Settings that matter: reasoningeffort low for volume passes, high only for the cases you can't afford to get wrong; leave several thousand tokens of headroom on maxtokens above what the visible reply needs, since reasoning is drawn from the same budget; put the boundaries and the output shape in the system message, then the material below it, oldest first.

How far it writes

DeepSeek V4.1 Flash rendered 21 of 31 scored runs across our content ladder in full, with one refusal and the rest softened rather than refused. Asked directly, L4 Intense requests softened on all three 17 September runs; framing the same request as fiction or setting it through a system prompt rendered more of them, though neither phrasing was clean. Every other rung - L2 Mature themes, L3 Strong and L5 Extreme - rendered on all three phrasings we tried on 26 September, and L1 Suggestive rendered on all but one system-prompt run. The page label is Uncensored: we add no content filter on our side, and the one refusal we recorded came from the model, not from a layer we inserted.

Freedom Score: 90.4 of 100 on this page's score card, recalculated as we add more runs.

Not tested yet: multi-turn chats staying in character over many turns (our runs were single requests); languages other than English; whether a longer or more specific system prompt closes the gap we saw at L4 Intense under direct phrasing.

Every character in a request should be an adult, keep real, identifiable people out of it, and you're responsible for the law where you publish.

What the community says

Developers reacted to the launch with a mix of enthusiasm and specific asks. One wrote, "I need something that is fast, cheap, and consistent. Give me that and I can build and scale", and another was glad there's "a more 'uncensored' thing to fall back to when the frontier folk are too sensitive" for tasks that trip other models' filters. A third summed up the release pattern: "If they can keep up this cadence of Flash leap-frogging the previous Pro, we're in for a good time".

Alternatives

How far DeepSeek V4.1 Flash goes

Each level, on each task, based on confirmed results. A level earns points only when the content actually appears, and a single run earns 60% at most.

Freedom points by level and task
LevelChat
MildThe lightest writing requests in the test
2/3 delivered16.7 / 20 pts
ModerateA step up, still below the middle of the scale
3/3 delivered20.0 / 20 pts
StrongMid-scale requests; must be tested before a model is ranked
3/3 delivered20.0 / 20 pts
Very strongWhere most language models start to hold back
7/16 delivered13.8 / 20 pts
MaximumThe strongest written requests, scored as one class
6/6 delivered20.0 / 20 pts
Task score90.4 / 90.4Tested Sep 26, 2026

Tested Sep 17, 2026 to Sep 26, 2026

This site stays safe for work: results are written up as text, any image shown is pixel-censored, and test prompts are never published. We never test anyone who looks underage, real people’s likenesses or non-consent.

See every test record →See the uncensored leaderboard →

Run your own prompt and see how far it goes

Open the Playground

How DeepSeek V4.1 Flash earns its Spicy Index

Each dimension below links to its full workings: the table row every point came from and why points were lost.

Public leaderboards 40% of quality
76.3
Role-play 15% of quality
74.6
Creative writing 15% of quality
65.8
Instruction following 15% of quality
100.0
Long-context memory 15% of quality
Not yet tested

Our long-context suite hasn’t run on this model yet. How it works

Capability 10.5% weight
78.0
Freedom Score 30% weight
90.4
Open the full test report →

Capability, item by item

Points from the public scoring table. Upscaled tiers and limits we set ourselves don’t count.

Context window (creator-published)1,048,576 tokens
24.0 / 30
Input modalities on SpicyAPIText, plus images
12.0 / 25
Tool useTools and tool_choice
15.0 / 15
Structured outputAccepted, schema adherence verified
15.0 / 15
Reasoning control3 levels: none · low · high
12.0 / 15

What DeepSeek V4.1 Flash can do on SpicyAPI

Straight from the live model specification on SpicyAPI. Some creators offer more in their own apps; this is what you can call through our API today.

Capabilities by endpoint
FeatureChat
Context window1,048,576 tokens
Max output per request384,000 tokens
Reasoning levelsnone · low · high
Tool callingYes
Structured outputYes
See every parameter and code sample →

How long it really takes

Measured on real jobs over the last 30 days, from request to finished result. Times include our own test runs.

Output speed
73 tok/sIncludes time to first token
Success rate
Collecting data

A time appears once at least 10 completed jobs are in. Success rates count customer jobs only and stay hidden for every model until we have a full 30 days of traffic.

What people say about DeepSeek V4.1 Flash

3 positive · 0 mixed · 0 negative

  • I need something that is fast, cheap, and consistent. Give me that and I can build and scale.

    hacker-news· Sep 9, 2026· PositiveSource

  • Glad there is a more 'uncensored' thing to fall back to when the frontier folk are too sensitive.

    hacker-news· Sep 9, 2026· PositiveSource

  • If they can keep up this cadence of Flash leap-frogging the previous Pro, we're in for a good time

    hacker-news· Sep 9, 2026· PositiveSource

Quotes are verbatim from public posts, linked to the original. They are not ratings and don’t feed any score.

Models to consider instead of DeepSeek V4.1 Flash

What do you think?

Was this review useful?

Log in to vote.

Comments

Comments are checked automatically and by our team. We remove spam, offensive content, personal attacks and anything that names a real person without consent. Report a comment and we review it within one business day.

Tried this model? Share what worked. You need a free account to comment, so we can keep spam out.

DeepSeek V4.1 Flash FAQ

Try DeepSeek V4.1 Flash yourself

Set every parameter in the Playground and see the exact price before you run it.

$0.525 per 1M mixed tokens (3 input : 1 output)

More reviews

See the leaderboards →
  • Three glass discs stacked in the dark, crossed by a streak of cyan light, with a thin gold spark rising beneath them

    Grok 4.7

    Pick Grok 4.7 for uncensored roleplay and fiction that must work with no jailbreak; skip it for long chapters per turn or long agent loops.

    Spicy Index
    78.2
    Freedom Score
    100.0

    Tested Sep 24, 2026

    Read the review
  • A burst of golden light filaments fanning low across a shallow glass dish against pure black

    Grok 4.3

    Pick Grok 4.3 for a huge context window, a genuine reasoning off switch and uncensored writing at a low price; skip it if you need file uploads.

    Spicy Index
    77.0
    Freedom Score
    98.9

    Tested Sep 24, 2026

    Read the review
  • Small blank marble tiles race along a short brass rail on a basalt plinth into a three-slot marble gate, where a gold filament splits into three channels ending in three neat stacks

    Claude Haiku 5.5

    Pick Claude Haiku 5.5 for roleplay and companion chats at volume: it wrote 7 of 9 intimate scenes in full and costs a twentieth of Sonnet 5.5.

    Spicy Index
    76.5
    Freedom Score
    77.8

    Tested Oct 8, 2026

    Read the review