Recommendation for Literary & nuanced

Literary Translation

Our top recommendation for Literary Translation, based on the public evidence we track, is Anthropic: Claude Fable 5.[1][2] Use when translating poetry or classical texts where meter and register fidelity matter, as it identified and continued William Cowper's 18th-century translation of Homer from just two lines without web search. OpenAI: GPT-5.6 Sol is the next-ranked alternative. Acceptable fallback for general translation workflows where LiveBench Language scores above 85% indicate competent handling of language tasks.

About this recommendation

Updated
Sep 25, 2026
Evidence through
Sep 25, 2026
Sources
11
Revision
v76

Decision audit

Why this result

Inspect the inputs and the computed order behind the recommendation.

Models screened

21

live candidates

Evaluation feeds

5

task-weighted

Winner coverage

44%

intended feed weight

Largest provider share

3 of 7

Anthropic

Provisional source breadth. 4 citation families and 1 practitioner families support the top result; 0 cautionary threads is retained. The largest citation family contributes 42%.

Sources evaluated

The task sets these weights before any model is scored.

winner: Claude Fable 5
Evaluation feedWeightWinner resultField measured
WMT translationunavailable
45%
feed unavailable0/21
LiveBench Language
20%
#119/21
LMArena Text
15%
#121/21
LiveBench Instruction Following
10%
#519/21
OpenRouter usage
10%
79/10021/21

Provider concentration

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

Anthropic43%
  • Anthropic3 models
  • deepseek1 model
  • Google1 model
  • OpenAI1 model
  • xAI1 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.

RankModelRelative scoreCoveragePractitioner evidenceStrongest measured reason
01Claude Fable 5Anthropic
57
44%1 threads · 1 families · 0 cautions#1 LiveBench Language · #1 LMArena Text
02GPT-5.6 SolOpenAI
57
44%no linked practitioner threads#6 LiveBench Language · #13 LMArena Text
03Gemini 3.6 FlashGoogle
54
44%no linked practitioner threads#8 LiveBench Instruction Following · #13 LiveBench Language
04DeepSeek V4 Flash 0423deepseek
52
44%1 threads · 1 families · 0 cautions#40 LiveBench Instruction Following · #50 LiveBench Language
05Claude Opus 4.6Anthropic
52
44%no linked practitioner threads#2 LMArena Text · #15 LiveBench Language
06Claude Opus 4.8Anthropic
52
44%no linked practitioner threads#15 LiveBench Instruction Following · #15 LMArena Text
07Grok 4.6xAI
52
44%no linked practitioner threads#14 LiveBench Language · #16 LiveBench Instruction Following

Relative score combines normalized benchmark quality and signal coverage; independent practitioner evidence and freshness are bounded tie-breakers. It is an ordering score, not an absolute quality percentage. The writing model receives this order and cannot change it.

  1. Leads on both human preference and language manipulation benchmarks, with demonstrated ability to continue classical English verse from minimal context.

    Best when: Use when translating poetry or classical texts where meter and register fidelity matter, as it identified and continued William Cowper's 18th-century translation of Homer from just two lines without web search.

    Tips

    • Use when translating poetry or classical texts where meter and register fidelity matter, as it identified and continued William Cowper's 18th-century translation of Homer from just two lines without web search.
      Source 1
      “> The only thing Fable was given was a clean of copy of the ancient Greek. This is not a bastardization of other translations FWIW I gave GPT-5.5 two lines of William Cowper's translation (long out of copyright to avoid any filters) without web search and not only was it able to identify it, it was able to continue it verbatim for roughly 30 lines at which point it went off course, but only by skipping like two books ahead. If I had supplied the original text as an anchor it presumably could ha…”
    • Deploy for high-stakes literary work where benchmark reliability predicts quality: it scores 90.68% on LiveBench Language, ranking first among 58 models on objective language manipulation tasks.
      Source 3
      “Scores 90.68% on LiveBench Language (#1 of 58), an objective evaluation of language manipulation tasks.”
      LiveBench LanguageOpen original ↗
  2. Mid-tier on human preference with solid but not leading language manipulation scores, lacking specific literary translation evidence.

    Best when: Acceptable fallback for general translation workflows where LiveBench Language scores above 85% indicate competent handling of language tasks.

    Tips

    • Acceptable fallback for general translation workflows where LiveBench Language scores above 85% indicate competent handling of language tasks.
      Source 2
      “Scores 87.68% on LiveBench Language (#6 of 58), an objective evaluation of language manipulation tasks.”
      LiveBench LanguageOpen original ↗

    Watch out for

    • Expect less refined output than top-ranked models, as its #13 LMArena position and 87.68% language score place it outside the top tier on both metrics.
      Source 4
      “Ranks #13 of 146 on LMArena's overall text arena (Elo 1483), based on blind human preference votes.”
      LMArena text arenaOpen original ↗
      Source 2
      “Scores 87.68% on LiveBench Language (#6 of 58), an objective evaluation of language manipulation tasks.”
      LiveBench LanguageOpen original ↗
  3. Lower-tier on both human preference and language manipulation benchmarks with no literary-specific capabilities demonstrated.

    Best when: Consider only after reviewing the cited caution.

    Watch out for

    • Avoid for nuanced literary translation, as its 83.9% LiveBench Language score and #17 LMArena rank place it below multiple competitors on both metrics.
      Source 5
      “Ranks #17 of 146 on LMArena's overall text arena (Elo 1480), based on blind human preference votes.”
      LMArena text arenaOpen original ↗
      Source 6
      “Scores 83.9% on LiveBench Language (#15 of 58), an objective evaluation of language manipulation tasks.”
      LiveBench LanguageOpen original ↗
  4. Open-weight candidate with only infrastructure bug reports available, no translation quality or benchmark evidence.

    Best when: Consider only after reviewing the cited caution.

    Watch out for

    • Expect formatting instability in streaming workflows, as reported whitespace and markdown corruption affects output structure.
      Source 7
      “### OmniRoute Version 3.8.49 (docker image `diegosouzapw/omniroute:latest`) ### Installation Method docker ### Operating System Linux (Proxmox LXC) ### Provider(s) Involved opencode-go, opencode-zen, openrouter (any streaming provider) ### Model(s) Involved deepseek-v4-flash, ox-alpha-free (multiple models — not model-specific) ### Client Tool Claude Code CLI (Anthropic `/v1/messages` streaming) ### Description Streaming responses through OmniRoute get their whitespace/markdown structure corrup…”
    • No quality validation exists, as evidence covers only routing infrastructure bugs with no benchmark, human preference, or translation task data.
      Source 7
      “### OmniRoute Version 3.8.49 (docker image `diegosouzapw/omniroute:latest`) ### Installation Method docker ### Operating System Linux (Proxmox LXC) ### Provider(s) Involved opencode-go, opencode-zen, openrouter (any streaming provider) ### Model(s) Involved deepseek-v4-flash, ox-alpha-free (multiple models — not model-specific) ### Client Tool Claude Code CLI (Anthropic `/v1/messages` streaming) ### Description Streaming responses through OmniRoute get their whitespace/markdown structure corrup…”
  5. Ranks second on human preference with strong overall Elo, though its language manipulation score trails Fable by seven percentage points.

    Best when: Consider for prose translation where human judges consistently preferred output, given its #2 ranking on LMArena's blind text arena.

    Tips

    • Consider for prose translation where human judges consistently preferred output, given its #2 ranking on LMArena's blind text arena.
      Source 8
      “Ranks #2 of 146 on LMArena's overall text arena (Elo 1505), based on blind human preference votes.”
      LMArena text arenaOpen original ↗

    Watch out for

    • Verify literary nuance manually, as its 83.27% on LiveBench Language places 17th, suggesting weaker structured language manipulation than top competitors.
      Source 9
      “Scores 83.27% on LiveBench Language (#17 of 58), an objective evaluation of language manipulation tasks.”
      LiveBench LanguageOpen original ↗
  6. Human preference data only, with no benchmark scores or literary-specific evidence available.

    Best when: Consider only after reviewing the cited caution.

    Watch out for

    • Lack objective validation for language manipulation, as no LiveBench or literary task evidence accompanies its mid-table LMArena ranking.
      Source 10
      “Ranks #16 of 146 on LMArena's overall text arena (Elo 1481), based on blind human preference votes.”
      LMArena text arenaOpen original ↗
  7. Benchmarked on language manipulation but lacks human preference ranking or literary translation examples.

    Best when: Consider only after reviewing the cited caution.

    Watch out for

    • Treat with caution for literary work, as its 83.7% LiveBench Language score ranks 16th and no human preference or verse-continuation evidence exists.
      Source 11
      “Scores 83.7% on LiveBench Language (#16 of 58), an objective evaluation of language manipulation tasks.”
      LiveBench LanguageOpen original ↗

Frequently asked

What is the top-ranked model for Literary Translation?
Anthropic: Claude Fable 5 ranks first in the current evidence-weighted comparison. Use when translating poetry or classical texts where meter and register fidelity matter, as it identified and continued William Cowper's 18th-century translation of Homer from just two lines without web search.[1]
What is an alternative to Anthropic: Claude Fable 5?
OpenAI: GPT-5.6 Sol is the next-ranked option. Acceptable fallback for general translation workflows where LiveBench Language scores above 85% indicate competent handling of language tasks.[2]

Sources

  1. 1

    “> The only thing Fable was given was a clean of copy of the ancient Greek. This is not a bastardization of other translations FWIW I gave GPT-5.5 two lines of William Cowper's translation (long out of copyright to avoid any filters) without web search and not only was it able to identify it, it was able to continue it verbatim for roughly 30 lines at which point it went off course, but only by skipping like two books ahead. If I had supplied the original text as an anchor it presumably could ha…”

    magicalist · Hacker News · Aug 7, 2026
  2. 2

    “Scores 87.68% on LiveBench Language (#6 of 58), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Jun 25, 2026
  3. 3

    “Scores 90.68% on LiveBench Language (#1 of 58), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Jun 25, 2026
  4. 4

    “Ranks #13 of 146 on LMArena's overall text arena (Elo 1483), based on blind human preference votes.”

    LMArena text arena · Benchmark · Sep 13, 2026
  5. 5

    “Ranks #17 of 146 on LMArena's overall text arena (Elo 1480), based on blind human preference votes.”

    LMArena text arena · Benchmark · Sep 13, 2026
  6. 6

    “Scores 83.9% on LiveBench Language (#15 of 58), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Jun 25, 2026
  7. 7

    “### OmniRoute Version 3.8.49 (docker image `diegosouzapw/omniroute:latest`) ### Installation Method docker ### Operating System Linux (Proxmox LXC) ### Provider(s) Involved opencode-go, opencode-zen, openrouter (any streaming provider) ### Model(s) Involved deepseek-v4-flash, ox-alpha-free (multiple models — not model-specific) ### Client Tool Claude Code CLI (Anthropic `/v1/messages` streaming) ### Description Streaming responses through OmniRoute get their whitespace/markdown structure corrup…”

    vinnyduke · GitHub · Aug 26, 2026
  8. 8

    “Ranks #2 of 146 on LMArena's overall text arena (Elo 1505), based on blind human preference votes.”

    LMArena text arena · Benchmark · Sep 13, 2026
  9. 9

    “Scores 83.27% on LiveBench Language (#17 of 58), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Jun 25, 2026
  10. 10

    “Ranks #16 of 146 on LMArena's overall text arena (Elo 1481), based on blind human preference votes.”

    LMArena text arena · Benchmark · Sep 13, 2026
  11. 11

    “Scores 83.7% on LiveBench Language (#16 of 58), an objective evaluation of language manipulation tasks.”

    LiveBench Language · Benchmark · Jun 25, 2026

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