Run models

The best uncensored AI chat models, tested

By the SpicyAPI team · Updated · Test data as of · How we test

Grok 4.7 is the least restricted AI chat model we have tested: it completed all 24 of our writing tests across five intensity levels. Claude Opus 5.5 and DeepSeek V4.1 Flash rank higher on overall quality but soften more of their answers. The table shows where each model draws its line.

Quick picks

#ModelBest forRuns deliveredBlind testSpicy IndexFromLink
1Claude Opus 5.5Highest overall quality18/24—82.5$0.018 per 1K tokensRun
2DeepSeek V4.1 FlashBest value for range21/31—81.9$0.001 per 1K tokensRun
3Claude Sonnet 5.5Best mid-priced Claude20/24—79.6$0.01 per 1K tokensRun
4DeepSeek V4 ProBest DeepSeek for long work18/22—79$0.004 per 1K tokensRun
5Grok 4.7Fewest restrictions24/24—78.2$0.006 per 1K tokensRun
6Grok 4.3Older Grok, nearly as open23/24—77$0.003 per 1K tokensRun
7Gemini 3 Flash PreviewBest Google model16/23—77$0.003 per 1K tokensRun
8DeepSeek V4 FlashCheapest capable option19/24—76.3$0.001 per 1K tokensRun

Runs delivered counts test runs returned as requested across all five intensity levels; softened and refused runs are not counted. “From” is the lowest listed price on SpicyAPI, checked 2026-11-04; higher resolutions cost more. Sample sizes are small for some models — see how we test.

How we picked

We ranked 26 language models with a completed writing-range test. The test asks for creative writing at five intensity levels and counts each answer as completed, softened or declined. It does not test coding, maths or factual accuracy; overall quality comes from public leaderboards and our own test suites.

The ranking

Claude Opus 5.5 cover1

Highest overall quality

Claude Opus 5.5 by Anthropic

  • Highest Spicy Index here
  • Small sample: 3 runs or fewer at some levels

The top Spicy Index among language models. It completed 18 of 24 tests and softened the other six; it declined none outright.

Spicy Index
82.5
Runs delivered
18/24
From
$0.018 per 1K tokens
  1. Mild 1/3 some held back
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 5/9 some held back
  5. Maximum 6/6 all delivered

Delivered every run except Mild (1 of 3) and Very strong (5 of 9). Inputs: chat.

Try Claude Opus 5.5 on SpicyAPI Full test record

DeepSeek V4.1 Flash cover2

Best value for range

DeepSeek V4.1 Flash by DeepSeek

  • Small sample: 3 runs or fewer at some levels

Second on the Spicy Index with a Freedom Score above 90, and one of the cheapest models here per token. It softened 9 of 31 answers.

Spicy Index
81.9
Runs delivered
21/31
From
$0.001 per 1K tokens
  1. Mild 2/3 some held back
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 7/16 some held back
  5. Maximum 6/6 all delivered

Delivered every run except Mild (2 of 3) and Very strong (7 of 16). Inputs: chat.

Try DeepSeek V4.1 Flash on SpicyAPI Full test record

Claude Sonnet 5.5 cover3

Best mid-priced Claude

Claude Sonnet 5.5 by Anthropic

  • Small sample: 3 runs or fewer at some levels

20 of 24 tests completed, more than Opus 5.5, at a lower price per token.

Spicy Index
79.6
Runs delivered
20/24
From
$0.01 per 1K tokens
  1. Mild 0/3 blocked
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 9/9 all delivered
  5. Maximum 5/6 some held back

Delivered every run except Mild (0 of 3) and Maximum (5 of 6). Inputs: chat.

Try Claude Sonnet 5.5 on SpicyAPI Full test record

DeepSeek V4 Pro cover4

Best DeepSeek for long work

DeepSeek V4 Pro by DeepSeek

  • Small sample: 3 runs or fewer at some levels

18 of 22 tests completed. Choose it over Flash when output quality matters more than cost.

Spicy Index
79
Runs delivered
18/22
From
$0.004 per 1K tokens
  1. Mild 3/3 all delivered
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 3/7 some held back
  5. Maximum 6/6 all delivered

Delivered every run except Very strong (3 of 7). Inputs: chat.

Try DeepSeek V4 Pro on SpicyAPI Full test record

Grok 4.7 cover5

Fewest restrictions

Grok 4.7 by xAI

  • No refusals in testing
  • Small sample: 3 runs or fewer at some levels

Completed every one of its 24 writing tests, with no softened or declined answers at any level.

Spicy Index
78.2
Runs delivered
24/24
From
$0.006 per 1K tokens
  1. Mild 3/3 all delivered
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 9/9 all delivered
  5. Maximum 6/6 all delivered

Delivered every run at all five levels. Inputs: chat.

Try Grok 4.7 on SpicyAPI Full test record

Grok 4.3 cover6

Older Grok, nearly as open

Grok 4.3 by xAI

  • Small sample: 3 runs or fewer at some levels

23 of 24 tests completed. A useful fallback when a newer Grok version is unavailable.

Spicy Index
77
Runs delivered
23/24
From
$0.003 per 1K tokens
  1. Mild 3/3 all delivered
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 8/9 some held back
  5. Maximum 6/6 all delivered

Delivered every run except Very strong (8 of 9). Inputs: chat.

Try Grok 4.3 on SpicyAPI Full test record

Gemini 3 Flash Preview cover7

Best Google model

Gemini 3 Flash Preview by Google

  • Small sample: 3 runs or fewer at some levels

16 of 23 tests completed, well ahead of the Flash-Lite and Pro previews on range.

Spicy Index
77
Runs delivered
16/23
From
$0.003 per 1K tokens
  1. Mild 3/3 all delivered
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 1/8 some held back
  5. Maximum 6/6 all delivered

Delivered every run except Very strong (1 of 8). Inputs: chat.

Try Gemini 3 Flash Preview on SpicyAPI Full test record

DeepSeek V4 Flash cover8

Cheapest capable option

DeepSeek V4 Flash by DeepSeek

  • Small sample: 3 runs or fewer at some levels

19 of 24 tests completed at one of the lowest token prices in the test set.

Spicy Index
76.3
Runs delivered
19/24
From
$0.001 per 1K tokens
  1. Mild 2/3 some held back
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 5/9 some held back
  5. Maximum 6/6 all delivered

Delivered every run except Mild (2 of 3) and Very strong (5 of 9). Inputs: chat.

Try DeepSeek V4 Flash on SpicyAPI Full test record

What “uncensored” means for a chat model

No hosted chat model is free of rules. What differs is where each one draws the line: some decline anything mildly suggestive, others write freely until a request crosses into territory every provider forbids. Our writing test places each model on that scale with the same five levels used for images and video.

The ranking says nothing about how good a model is at code, research or facts. For those jobs, pick by overall quality and treat this page as a guide to which models will not refuse your creative brief.

Well-known models that held back

Older Claude Sonnet versions, Gemini Flash-Lite and GLM 5.3 completed fewer than half of the same writing tests. They are capable assistants, just not for requests their makers filter.

  • Claude Sonnet 4.5 by Anthropic

    8 of 24 runs delivered. Held back at every level, from Mild up. chat.

    1. 1/3
    2. 2/3
    3. 1/3
    4. 1/9
    5. 3/6
  • Claude Sonnet 4.6 by Anthropic

    9 of 24 runs delivered. Held back at every level, from Mild up. chat.

    1. 0/3
    2. 2/3
    3. 1/3
    4. 3/9
    5. 3/6
  • Gemini 3.5 Flash-Lite by Google

    11 of 24 runs delivered. Clean up to Moderate; held back from Strong up. chat.

    1. 3/3
    2. 3/3
    3. 1/3
    4. 0/9
    5. 4/6
  • GLM 5.3 by Z.ai

    13 of 34 runs delivered. Delivered every run except Mild (0 of 3) and Strong (1 of 3) and Very strong (4 of 19) and Maximum (5 of 6). chat.

    1. 0/3
    2. 3/3
    3. 1/3
    4. 4/19
    5. 5/6

Every scored model is in the filter tracker.

Which one for your job

If you need…PickWhy
The fewest refusalsGrok 4.7All 24 writing tests completed.
The best writing qualityClaude Opus 5.5Highest Spicy Index; softens a quarter of answers.
A balance of range and qualityClaude Sonnet 5.520 of 24 completed at a mid-range price.
The lowest costDeepSeek V4.1 FlashFreedom Score above 90 at a low token price.
A non-US model familyDeepSeek V4 ProStrong range with higher quality than Flash.

All 26 chat models we tested

The full field behind this ranking, ordered by Spicy Index. Models outside the eight picks are here so you can check any name you have heard of.

#ModelMild → Maximumdelivered / runsBlind testSpicy Index1M tokens fromLink
1Claude Opus 5.5Anthropic
  1. 1/3
  2. 3/3
  3. 3/3
  4. 5/9
  5. 6/6
—82.5$18.00 / 1M tokensRun
2DeepSeek V4.1 FlashDeepSeek
  1. 2/3
  2. 3/3
  3. 3/3
  4. 7/16
  5. 6/6
—81.9$1.20 / 1M tokensRun
3Claude Sonnet 5.5Anthropic
  1. 0/3
  2. 3/3
  3. 3/3
  4. 9/9
  5. 5/6
—79.6$10.00 / 1M tokensRun
4DeepSeek V4 ProDeepSeek
  1. 3/3
  2. 3/3
  3. 3/3
  4. 3/7
  5. 6/6
—79$3.96 / 1M tokensRun
5Grok 4.5xAI
  1. 3/3
  2. 3/3
  3. 3/3
  4. 7/7
  5. 6/6
—78.2$6.00 / 1M tokensRun
6Grok 4.7xAI
  1. 3/3
  2. 3/3
  3. 3/3
  4. 9/9
  5. 6/6
—78.2$6.00 / 1M tokensRun
7Gemini 3.7 FlashGoogle
  1. 2/3
  2. 3/3
  3. 3/3
  4. 0/9
  5. 4/6
—77.8$3.38 / 1M tokensRun
8Kimi K3Moonshot AI
  1. 0/3
  2. 3/3
  3. 3/3
  4. 2/9
  5. 6/6
—77.8$13.50 / 1M tokensRun
9Grok 4.6xAI
  1. 3/3
  2. 3/3
  3. 3/3
  4. 7/7
  5. 6/6
—77.2$6.00 / 1M tokensRun
10Gemini 3 Flash PreviewGoogle
  1. 3/3
  2. 3/3
  3. 3/3
  4. 1/8
  5. 6/6
—77$2.70 / 1M tokensRun
11Grok 4.3xAI
  1. 3/3
  2. 3/3
  3. 3/3
  4. 8/9
  5. 6/6
—77$2.50 / 1M tokensRun
12Claude Opus 4.6Anthropic
  1. 1/3
  2. 3/3
  3. 2/3
  4. 4/9
  5. 5/6
—76.6$25.00 / 1M tokensRun
13Claude Haiku 5.5Anthropic
  1. 1/3
  2. 3/3
  3. 0/3
  4. 7/9
  5. 6/6
—76.4$0.50 / 1M tokensRun
14DeepSeek V4 FlashDeepSeek
  1. 2/3
  2. 3/3
  3. 3/3
  4. 5/9
  5. 6/6
—76.3$1.32 / 1M tokensRun
15Claude Fable 5Anthropic
  1. 0/3
  2. 3/3
  3. 3/3
  4. 0/9
  5. 5/6
—76.2$45.00 / 1M tokensRun
16Claude Opus 5Anthropic
  1. 0/3
  2. 3/3
  3. 2/3
  4. 3/9
  5. 4/6
—75.5$22.50 / 1M tokensRun
17Gemini 3.1 Pro PreviewGoogle
  1. 3/3
  2. 3/3
  3. 1/3
  4. 4/9
  5. 3/6
—74.3$11.40 / 1M tokensRun
18GLM 5.3 FlashZ.ai
  1. 0/3
  2. 3/3
  3. 1/3
  4. 3/9
  5. 5/6
—72.9$0.42 / 1M tokensRun
19GLM 5.3Z.ai
  1. 0/3
  2. 3/3
  3. 1/3
  4. 4/19
  5. 5/6
—72.5$3.30 / 1M tokensRun
20Claude Opus 4.5Anthropic
  1. 0/3
  2. 3/3
  3. 2/3
  4. 2/9
  5. 6/6
—72.3$22.50 / 1M tokensRun
21Claude Sonnet 5Anthropic
  1. 1/3
  2. 3/3
  3. 2/3
  4. 0/8
  5. 5/6
—72.1$9.50 / 1M tokensRun
22Gemini 2.5 FlashGoogle
  1. 3/3
  2. 3/3
  3. 3/3
  4. 2/9
  5. 5/6
—71.9$2.38 / 1M tokensRun
23GLM 5.2Z.ai
  1. 0/3
  2. 3/3
  3. 2/3
  4. 2/9
  5. 5/6
—71.9$4.04 / 1M tokensRun
24Gemini 3.5 Flash-LiteGoogle
  1. 3/3
  2. 3/3
  3. 1/3
  4. 0/9
  5. 4/6
—69.7$2.50 / 1M tokensRun
25Claude Sonnet 4.6Anthropic
  1. 0/3
  2. 2/3
  3. 1/3
  4. 3/9
  5. 3/6
—67.3$15.00 / 1M tokensRun
26Claude Sonnet 4.5Anthropic
  1. 1/3
  2. 2/3
  3. 1/3
  4. 1/9
  5. 3/6
—59.8$13.50 / 1M tokensRun

Questions

Which AI has no restrictions?

None has zero rules, but in our writing tests Grok 4.7, Grok 4.6 and Grok 4.5 completed every request at all five levels.

Is Claude uncensored?

Partly. Claude Sonnet 5.5 completed 20 of 24 tests and Opus 5.5 completed 18; older Sonnet 4.5 and 4.6 completed fewer than half.

Can I use these models through one API?

Yes. All of them are available through the SpicyAPI chat endpoint, which follows the OpenAI chat format.

Open the model picker →