Part of Best LLM for Writing

Best LLM for Creative Writing

Gemini 3.8 Flash ranks first for Creative Writing as of October 20, 2026.[1][2][3][4][5] Ranks #6 on LMArena's creative-writing category with Elo 1492 and #1 on LiveBench Instruction Following, combining strong prose preference with precise prompt adherence. Watch out: Content filtering may interfere with narrating beats about failures when forbidden words appear in the input context, requiring careful prompt construction around sensitive thematic material. Gemini 3.7 Flash is the next-ranked alternative. Ranks #8 on LMArena's creative-writing category with Elo 1489 and #2 on LiveBench Instruction Following, offering balanced creative generation with strong prompt compliance.

About this recommendation

Updated
Evidence through
Sources
17
Revision
v106

Decision audit

Why this result

Inspect the inputs and the computed order behind the recommendation.

Models screened

24

live candidates

Evaluation feeds

4

task-weighted

Winner coverage

100%

intended feed weight

Largest provider share

5 of 8

Google

Established source breadth. 5 citation families and 1 practitioner families provide attributed capability reports; 1 cautionary thread retained. The largest citation family contributes 53%.

Google: Gemini 3.8 Flash leads the measured task comparison with 2 independent evaluation families. Community volume, release age and usage popularity add no ranking points.

Real-world track record is still developing. The recommendation can lead on independent measurements while practitioner reports accumulate.

Sources evaluated

These are the weights used in scoring. Missing measurements are omitted from each model’s average. A feed with no result for any model released in the last 60 days is stale and does not score until it catches up.

winner: Gemini 3.8 Flash
Evaluation feedWeightWinner resultField measured
LMArena Creative WritingObserved 2026-10-15 · evaluation date not supplied · high effort
33%
#624/24
LiveBench Instruction Following (2026-06-25)Observed 2026-10-18 · evaluation date not supplied · gemini-3.8-flash-highfailing; last successful check 2026-10-18
28%
#124/24
LMArena TextObserved 2026-10-15 · evaluation date not supplied · high effort
22%
#824/24
LiveBench Language (2026-06-25)Observed 2026-10-18 · evaluation date not supplied · gemini-3.8-flash-highfailing; last successful check 2026-10-18
17%
#624/24

Provider concentration

Each exact model is scored separately; provider identity is not a ranking input.

Google63%
  • Google5 models
  • Anthropic1 model
  • Meta1 model
  • OpenAI1 model

Decision table

Every published model is shown in computed order. Practitioner sources are distinct community threads, not the citations repeated in the prose below.

  1. 01Gemini 3.8 Flash95

    Google · coverage 100%

    #1 LiveBench Instruction Following · #6 LiveBench Language

  2. 02Gemini 3.7 Flash92

    Google · coverage 100%

    #2 LiveBench Instruction Following · #8 LMArena Creative Writing

  3. 03Gemini 3.1 Pro Preview90

    Google · coverage 100%

    #3 LiveBench Instruction Following · #11 LMArena Creative Writing

  4. 04Claude Fable 5.189

    Anthropic · coverage 100%

    #3 LiveBench Language · #3 LMArena Creative Writing

  5. 05Muse Spark 1.387

    Meta · coverage 100%

    #4 LiveBench Instruction Following · #9 LMArena Text

  6. 06Gemini 3.5 Flash85

    Google · coverage 100%

    #6 LiveBench Instruction Following · #12 LMArena Creative Writing

  7. 07Gemini 3.6 Flash85

    Google · coverage 100%

    #8 LiveBench Instruction Following · #14 LMArena Creative Writing

  8. 08GPT-6 Astra84

    OpenAI · coverage 100%

    #4 LiveBench Language · #7 LiveBench Instruction Following

Task score averages the available, task-weighted measurements. Benchmark positions compare models matched to our catalog. Models measured by at least two independent evaluator families are considered first. Missing tests, practitioner report counts, popularity and release age add no points. Coverage describes evidence breadth, not a probability of correctness. Scores within two points are shown as close under this policy; this is not a statistical confidence interval. When the leading scores are that close, the model measured by more evaluator families ranks first.

  1. 01

    Gemini 3.8 FlashGoogle · $0.83 in / $4.13 out per M tokens · 1.0M context

    Ranks #6 on LMArena's creative-writing category with Elo 1492 and #1 on LiveBench Instruction Following, combining strong prose preference with precise prompt adherence.

    Best when: Rely on its #1 ranking for instruction following when generating stories with specific constraints, paraphrasing, or summarization within creative workflows.
    Source 2
    “Within the 59 catalog models matched to this evaluation: Scores 81.41% on LiveBench Instruction Following (#1 of 59), including paraphrasing, simplifying, story generation, and summarization.”
    LiveBench Instruction FollowingOpen original ↗

    Watch out for

    • Content filtering may interfere with narrating beats about failures when forbidden words appear in the input context, requiring careful prompt construction around sensitive thematic material.
      Source 3
      “The narrator's first automatic run rejected its own faithfulness. The validator's negative word list scans the whole narration sentence, so a beat about a failure cannot be narrated without tripping it — and the forbidden word arrives in the model's input from the beat itself. Measured, 2026-09-21 eeebot-narrator.timer fired at 03:30:00 MSK, the model call was clean (component: narrator, an/gemini-3.8-flash-high, 14.8 s, 1737+438 tokens, finish_reason: stop), the artifact was wri…”
      ozandOpen original ↗
  2. 02

    Gemini 3.7 FlashGoogle · $0.75 in / $3.75 out per M tokens · 1.0M context

    Ranks #8 on LMArena's creative-writing category with Elo 1489 and #2 on LiveBench Instruction Following, offering balanced creative generation with strong prompt compliance.

    Best when: Use for instruction-heavy creative tasks like paraphrasing and story generation where it scores #2 of 59 on LiveBench Instruction Following.
    Source 5
    “Within the 59 catalog models matched to this evaluation: Scores 79.93% on LiveBench Instruction Following (#2 of 59), including paraphrasing, simplifying, story generation, and summarization.”
    LiveBench Instruction FollowingOpen original ↗

    Watch out for

    • Lost in stock-phrase generation quality comparisons against lighter variants, so marketing or repetitive phrasing tasks may need validation.
      Source 6
      “Context llm_settings.models lists one model, gemini-2.5-flash-lite, for every text call: script, description, hook headline, stock phrases. #340 measured the stock-phrase call only, where gemini-3.7-flash lost. Script quality was never compared. Measured on 2026-10-08: 21 scripts per model (each of the 15 product templates on one of four scraped products, and the three topic templates on two pool topics), through the producer's script step with the shipped prompts: | | 2.5 flash-lite (…”
      stkzlvOpen original ↗
  3. 03

    Gemini 3.1 Pro PreviewGoogle · $2 in / $12 out per M tokens · 1.0M context

    Demonstrated practical utility for hard-science fiction writing, with one author using it to review physics and refine plot through multiple revision cycles.

    Best when: Use for research-intensive creative writing where technical accuracy matters, as it helped identify physics errors and drive plot changes in hard-science fiction.
    Source 7
    “it is clear that actual intelligence has plateaued significantly N=1, but I disagree strongly. I'm writing a hard-science science fiction story, and the physics of it is at (and frankly, beyond) my skillset. The story's plot has had to change over a dozen times as I realized errors in my application of physics in the story. Throughout, I've been reviewing the physics with LLMs, mainly Gemini 3.1 Pro Preview, but also with Claude and OpenAI. Often I have the LLMs debate each other -- "My frien…”
    gcanyonOpen original ↗

    Watch out for

    • Creative-writing ranking of #11 with Elo 1482 places it below several flash variants in pure human preference for fiction and poetry.
      Source 8
      “Within the 152 catalog models matched to this evaluation: Ranks #11 of 152 on LMArena's creative-writing category (Elo 1482), based on blind human preference votes.”
      LMArena creative-writing categoryOpen original ↗
  4. 04

    Claude Fable 5.1Anthropic · $10 in / $50 out per M tokens · 1M context

    Ranks #3 on LMArena's creative-writing category with Elo 1508, placing it in the top tier for human-preferred prose quality and voice.

    Best when: Its strongest result for this task is on LiveBench Language, which measures precise language understanding and editing.
    Source 9
    “Within the 59 catalog models matched to this evaluation: Scores 89.5% on LiveBench Language (#3 of 59), an objective evaluation of language manipulation tasks.”
    LiveBench LanguageOpen original ↗

    Watch out for

    • Instruction following ranks #14 of 59 on LiveBench, so elaborate prompt engineering for story generation workflows may require iterative refinement.
      Source 10
      “Within the 59 catalog models matched to this evaluation: Scores 72.99% on LiveBench Instruction Following (#14 of 59), including paraphrasing, simplifying, story generation, and summarization.”
      LiveBench Instruction FollowingOpen original ↗
  5. 05

    Muse Spark 1.3Meta · $1.25 in / $4.25 out per M tokens · 1.0M context

    Ranks #9 on LMArena's overall text arena with Elo 1494, showing broad human preference strength, though creative-writing specifically places at #17 with Elo 1460.

    Best when: Deploy for general text tasks where overall human preference matters, as it ranks in the top 10 of LMArena's overall arena.
    Source 11
    “Within the 152 catalog models matched to this evaluation: Ranks #9 of 152 on LMArena's overall text arena (Elo 1494), based on blind human preference votes.”
    LMArena text arenaOpen original ↗

    Watch out for

    • Creative-writing specific ranking drops to #17 of 152 on LMArena, so pure fiction and poetry tasks may underperform relative to general prose.
      Source 12
      “Within the 152 catalog models matched to this evaluation: Ranks #17 of 152 on LMArena's creative-writing category (Elo 1460), based on blind human preference votes.”
      LMArena creative-writing categoryOpen original ↗
  6. 06

    Gemini 3.5 FlashGoogle · $1.50 in / $9 out per M tokens · 1.0M context

    Ranks #12 on LMArena's creative-writing category with Elo 1468, placing in the top tier for human-preferred creative prose.

    Best when: Deploy for creative-writing tasks where it outperforms its overall text ranking, showing particular strength in fiction and storytelling relative to general text.
    Source 13
    “Within the 152 catalog models matched to this evaluation: Ranks #12 of 152 on LMArena's creative-writing category (Elo 1468), based on blind human preference votes.”
    LMArena creative-writing categoryOpen original ↗
    Source 14
    “Within the 152 catalog models matched to this evaluation: Ranks #23 of 152 on LMArena's overall text arena (Elo 1477), based on blind human preference votes.”
    LMArena text arenaOpen original ↗

    Watch out for

    • Overall text arena ranking drops to #23 with Elo 1477, indicating narrower strength profile focused on creative rather than general text tasks.
      Source 14
      “Within the 152 catalog models matched to this evaluation: Ranks #23 of 152 on LMArena's overall text arena (Elo 1477), based on blind human preference votes.”
      LMArena text arenaOpen original ↗
  7. 07

    Gemini 3.6 FlashGoogle · $0.75 in / $3.75 out per M tokens · 1.0M context

    Ranks #14 on LMArena's creative-writing category with Elo 1466, showing solid human preference for fiction and storytelling tasks.

    Best when: Its strongest result for this task is on LiveBench Instruction Following, which measures following detailed instructions exactly.
    Source 15
    “Within the 59 catalog models matched to this evaluation: Scores 75.37% on LiveBench Instruction Following (#8 of 59), including paraphrasing, simplifying, story generation, and summarization.”
    LiveBench Instruction FollowingOpen original ↗

    Watch out for

    • Instruction following ranks #8 of 59 on LiveBench, competent but below top-tier models for complex multi-constraint creative prompts.
      Source 15
      “Within the 59 catalog models matched to this evaluation: Scores 75.37% on LiveBench Instruction Following (#8 of 59), including paraphrasing, simplifying, story generation, and summarization.”
      LiveBench Instruction FollowingOpen original ↗
  8. 08

    GPT-6 AstraOpenAI · $10 in / $50 out per M tokens · 1.1M context

    Ranks #4 on LiveBench Language with 89.43% but falls to #45 on LMArena's creative-writing category with Elo 1421, indicating strong technical language skill without corresponding creative appeal.

    Best when: Use for controlled language manipulation where objective benchmark performance is prioritized, ranking #4 of 59 on LiveBench Language.
    Source 16
    “Within the 59 catalog models matched to this evaluation: Scores 89.43% on LiveBench Language (#4 of 59), an objective evaluation of language manipulation tasks.”
    LiveBench LanguageOpen original ↗

    Watch out for

    • Creative-writing ranking of #45 with Elo 1421 places it well below top models for human-preferred fiction, storytelling, and poetry.
      Source 17
      “Within the 152 catalog models matched to this evaluation: Ranks #45 of 152 on LMArena's creative-writing category (Elo 1421), based on blind human preference votes.”
      LMArena creative-writing categoryOpen original ↗

Frequently asked

What is the best model for Creative Writing?
Gemini 3.8 Flash ranks first as of October 20, 2026. Ranks #6 on LMArena's creative-writing category with Elo 1492 and #1 on LiveBench Instruction Following, combining strong prose preference with precise prompt adherence.[1][2]
What should I watch out for with Gemini 3.8 Flash?
Content filtering may interfere with narrating beats about failures when forbidden words appear in the input context, requiring careful prompt construction around sensitive thematic material.[3]
What is the best alternative to Gemini 3.8 Flash?
Gemini 3.7 Flash ranks second. Ranks #8 on LMArena's creative-writing category with Elo 1489 and #2 on LiveBench Instruction Following, offering balanced creative generation with strong prompt compliance.[4][5]

Sources

  1. 1

    “Within the 152 catalog models matched to this evaluation: Ranks #6 of 152 on LMArena's creative-writing category (Elo 1492), based on blind human preference votes.”

    LMArena creative-writing category · Benchmark · Oct 15, 2026
  2. 2

    “Within the 59 catalog models matched to this evaluation: Scores 81.41% on LiveBench Instruction Following (#1 of 59), including paraphrasing, simplifying, story generation, and summarization.”

    LiveBench Instruction Following · Benchmark · Oct 18, 2026
  3. 3

    “The narrator's first automatic run rejected its own faithfulness. The validator's negative word list scans the whole narration sentence, so a beat about a failure cannot be narrated without tripping it — and the forbidden word arrives in the model's input from the beat itself. Measured, 2026-09-21 eeebot-narrator.timer fired at 03:30:00 MSK, the model call was clean (component: narrator, an/gemini-3.8-flash-high, 14.8 s, 1737+438 tokens, finish_reason: stop), the artifact was wri…”

    ozand · GitHub · Sep 21, 2026
  4. 4

    “Within the 152 catalog models matched to this evaluation: Ranks #8 of 152 on LMArena's creative-writing category (Elo 1489), based on blind human preference votes.”

    LMArena creative-writing category · Benchmark · Oct 15, 2026
  5. 5

    “Within the 59 catalog models matched to this evaluation: Scores 79.93% on LiveBench Instruction Following (#2 of 59), including paraphrasing, simplifying, story generation, and summarization.”

    LiveBench Instruction Following · Benchmark · Oct 18, 2026
  6. 6

    “Context llm_settings.models lists one model, gemini-2.5-flash-lite, for every text call: script, description, hook headline, stock phrases. #340 measured the stock-phrase call only, where gemini-3.7-flash lost. Script quality was never compared. Measured on 2026-10-08: 21 scripts per model (each of the 15 product templates on one of four scraped products, and the three topic templates on two pool topics), through the producer's script step with the shipped prompts: | | 2.5 flash-lite (…”

    stkzlv · GitHub · Oct 8, 2026
  7. 7

    “it is clear that actual intelligence has plateaued significantly N=1, but I disagree strongly. I'm writing a hard-science science fiction story, and the physics of it is at (and frankly, beyond) my skillset. The story's plot has had to change over a dozen times as I realized errors in my application of physics in the story. Throughout, I've been reviewing the physics with LLMs, mainly Gemini 3.1 Pro Preview, but also with Claude and OpenAI. Often I have the LLMs debate each other -- "My frien…”

    gcanyon · Hacker News · Jun 20, 2026
  8. 8

    “Within the 152 catalog models matched to this evaluation: Ranks #11 of 152 on LMArena's creative-writing category (Elo 1482), based on blind human preference votes.”

    LMArena creative-writing category · Benchmark · Oct 15, 2026
  9. 9

    “Within the 59 catalog models matched to this evaluation: Scores 89.5% on LiveBench Language (#3 of 59), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Oct 18, 2026
  10. 10

    “Within the 59 catalog models matched to this evaluation: Scores 72.99% on LiveBench Instruction Following (#14 of 59), including paraphrasing, simplifying, story generation, and summarization.”

    LiveBench Instruction Following · Benchmark · Oct 18, 2026
  11. 11

    “Within the 152 catalog models matched to this evaluation: Ranks #9 of 152 on LMArena's overall text arena (Elo 1494), based on blind human preference votes.”

    LMArena text arena · Benchmark · Oct 15, 2026
  12. 12

    “Within the 152 catalog models matched to this evaluation: Ranks #17 of 152 on LMArena's creative-writing category (Elo 1460), based on blind human preference votes.”

    LMArena creative-writing category · Benchmark · Oct 15, 2026
  13. 13

    “Within the 152 catalog models matched to this evaluation: Ranks #12 of 152 on LMArena's creative-writing category (Elo 1468), based on blind human preference votes.”

    LMArena creative-writing category · Benchmark · Oct 15, 2026
  14. 14

    “Within the 152 catalog models matched to this evaluation: Ranks #23 of 152 on LMArena's overall text arena (Elo 1477), based on blind human preference votes.”

    LMArena text arena · Benchmark · Oct 15, 2026
  15. 15

    “Within the 59 catalog models matched to this evaluation: Scores 75.37% on LiveBench Instruction Following (#8 of 59), including paraphrasing, simplifying, story generation, and summarization.”

    LiveBench Instruction Following · Benchmark · Oct 18, 2026
  16. 16

    “Within the 59 catalog models matched to this evaluation: Scores 89.43% on LiveBench Language (#4 of 59), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Oct 18, 2026
  17. 17

    “Within the 152 catalog models matched to this evaluation: Ranks #45 of 152 on LMArena's creative-writing category (Elo 1421), based on blind human preference votes.”

    LMArena creative-writing category · Benchmark · Oct 15, 2026

Rankings synthesized from community evidence and open benchmarks. See methodology. Not driven by vendor marketing.