Best LLM for Writing (July 2026): Fable 5 vs GPT-5.6 Tested

As of July 2026, Claude Fable 5 is the best LLM for writing on BenchLM's board: it holds the highest Arena Elo (1508) of any tracked model and Claude's traditional instruction-following edge, at $10/$50 per million tokens. GPT-5.6 Sol, launched July 9, is the main new challenger and is being benchmarked now. Gemini 3.1 Pro remains the value pick.

The detailed rankings below reflect our full creative-writing and instruction-following test set, where Claude Opus 4.6 led the previous generation for long-form content and Gemini 3.1 Pro led raw creative-writing scores at a materially cheaper token cost. The same decision logic applies tier-for-tier to the current lineup.

Writing quality is harder to benchmark than coding or math. There's no SWE-bench equivalent for prose: no single score that tells you which model writes the best blog post. Instead, we use a combination of Arena creative writing Elo (crowd-sourced human preference), instruction-following benchmarks (IFEval), and knowledge scores that affect factual accuracy.

Top writing models, ranked

Table 1 Download CSV

Model	Arena Creative Writing	Arena Instruction Following	IFEval	MMLU	Price (in/out)
Gemini 3.1 Pro	1487	1490	95	99	$2/$12
Claude Opus 4.6	1468	1500	95	99	$5/$25
GPT-5.4 Pro	1461	1488	97	99	$30/$180
Claude Sonnet 4.6	1443	1479	89.5	99	$3/$15
GLM-5 (Reasoning)	1442	1445	92	96	—
Grok 4.1	1431	1433	93	99	—
GPT-5.4	1423	1470	96	99	$2.50/$15

Scores from BenchLM.ai. Arena Elo from arena.ai. Prices per million tokens.

Two metrics matter most for writing: Arena Creative Writing measures whether humans prefer one model's prose over another in blind comparisons. IFEval measures whether a model follows specific formatting and style instructions, critical for writers who need a particular tone, structure, or length.

Claude Opus 4.6: the best writing model in 2026

Claude Opus 4.6 isn't the highest on Arena creative writing (Gemini 3.1 Pro leads by 18 Elo points). But it leads on instruction following, both on Arena's human-preference IF score (1500) and on IFEval (95).

Why does instruction following matter more than raw creative writing for most writers? Because real writing work isn't "write me something creative." It's "match this brand voice, keep it under 800 words, use this structure, don't use these phrases." That's instruction following.

Claude's non-reasoning architecture is also an advantage for writing. Reasoning models (GPT-5.4 Pro, GLM-5 Reasoning) pause to "think" before responding, which adds latency and can produce overly analytical prose. Claude generates naturally, better for iterative drafting where you go back and forth refining tone and structure.

At $5/$25 per million tokens, Claude Opus 4.6 is still premium-priced. For professional writers and content teams where quality directly drives revenue, the premium is justified. For everyone else, keep reading.

Best for specific writing tasks

Blog posts and long-form articles

Long-form content demands consistent voice across thousands of words, accurate claims, and good structure. Instruction following and knowledge benchmarks both matter here.

Best option: Claude Opus 4.6, highest Arena IF (1500), strong knowledge scores (MMLU: 99, GPQA: 91.3, HLE: 53), and produces coherent long-form prose without drifting. Its 1M context window handles long outlines and reference material.

Best value: Gemini 3.1 Pro, Arena CW: 1487 (highest), IFEval: 95, MMLU: 99, GPQA: 97. At $2/$12, it is still cheaper than the flagship premium tier and also has a 1M context window.

Copywriting and marketing

Short-form copy needs to be punchy, conversion-oriented, and brand-consistent. Instruction following matters most: you need the model to nail your tone guidelines on the first try.

Best option: GPT-5.4, IFEval: 96, strong structured output. Excels at ad copy, landing pages, and email sequences where you need a specific format and call-to-action pattern.

Best value: Claude Sonnet 4.6, Arena IF: 1479, IFEval: 89.5. At $3/$15, it's 5x cheaper than Opus with roughly 90% of the writing quality. Good enough for most marketing copy.

Email newsletters and outreach

Volume matters for email. You're writing dozens or hundreds of variations, not one perfect piece.

Best option: Gemini 3.1 Pro, the highest creative writing score at a relatively affordable frontier price. $2/$12 still makes batch generation practical.

Budget option: Gemini 3 Flash, Arena CW: 1461 at $0.50/$3. For high-volume outreach where you test many variants, Flash delivers roughly 85% of Pro quality at 40% of the cost.

Fiction and creative writing

Fiction is the one writing task where Arena creative writing Elo matters most. You want imagination, voice, and surprise, not merely instruction compliance.

Best option: Gemini 3.1 Pro, leads with 1487 on Arena creative writing. Strong at maintaining character voice and narrative consistency across long outputs.

Runner-up: Claude Opus 4.6, 1468 Arena CW. Many fiction writers prefer Claude's prose style despite the slightly lower creative writing Elo, particularly for literary fiction and editing.

Editing and rewriting

Editing requires the model to understand your intent without overwriting your voice. Instruction following is paramount: the model needs to change what you asked and leave everything else alone.

Best option: Claude Opus 4.6, Arena IF: 1500 (highest). Its tendency to follow instructions precisely makes it the most reliable editor. It's less likely to "improve" things you didn't ask it to change.

Budget option: GPT-5.4, IFEval: 96, Arena IF: 1470. Cheaper at $2.50/$15 and still strong at targeted edits.

The benchmarks that matter for writing

Arena creative writing (Elo)

Chatbot Arena runs blind head-to-head comparisons where humans pick which response they prefer. The creative writing category specifically tests prose quality, storytelling, and stylistic range. It's the closest thing we have to a human preference benchmark for writing.

Limitation: Arena measures which model humans prefer in short comparisons, not which produces the best 2,000-word blog post. Short-form preference doesn't always translate to long-form quality.

IFEval (instruction following)

IFEval measures whether a model follows specific verifiable instructions: "write exactly 3 paragraphs," "don't use the word 'innovative,'" "respond in all caps." This directly maps to real writing workflows where you need format and style constraints followed precisely.

MMLU and knowledge benchmarks

Writing quality depends partly on factual accuracy. Models with stronger knowledge benchmarks (MMLU, GPQA) produce fewer factual errors in informational content. The gap is smallest at the frontier (all top models score 96+ on MMLU) but becomes significant at lower price tiers.

Top 5 models: detailed breakdown

1. Claude Opus 4.6: best for professional writers

Table 2 Download CSV

Arena Creative Writing	1468
Arena Instruction Following	1500
IFEval	95
Price	$5/$25 per million tokens
Context	1M tokens

Pros: Highest instruction-following scores across both Arena and IFEval. Non-reasoning architecture produces natural, fluid prose. Excels at editing: changes what you ask without overwriting your voice. Strong knowledge base (HLE: 53, highest among writing models).

Cons: Premium-priced frontier model at $5/$25. Overkill for simple writing tasks. Arena creative writing score (1468) is below Gemini 3.1 Pro.

Best for: Professional content teams, book editing, brand-voice-sensitive copy, long-form journalism.

2. Gemini 3.1 Pro: best value for writing

Table 3 Download CSV

Arena Creative Writing	1487
Arena Instruction Following	1490
IFEval	95
Price	$1.25/$5 per million tokens
Context	1M tokens

Pros: Highest Arena creative writing score. Matches Claude Opus on IFEval (both 95). 12x cheaper on input than Claude Opus, 2x cheaper than GPT-5.4. 1M context window handles massive documents.

Cons: Prose style can feel less distinctive than Claude's. GPQA: 97 and MMLU: 99 are strong but the writing "feel" is more functional than literary.

Best for: Content marketers, bloggers, email marketers, fiction writers, anyone who values quality-per-dollar.

3. GPT-5.4: best for structured content

Table 4 Download CSV

Arena Creative Writing	1423
Arena Instruction Following	1470
IFEval	96
Price	$2.50/$15 per million tokens
Context	1.05M tokens

Pros: Highest IFEval score among non-Pro models (96). Strong at structured, analytical writing: whitepapers, technical docs, report generation. Excellent knowledge scores (GPQA: 92.8, MMLU: 99). Familiar ChatGPT interface.

Cons: Lower Arena creative writing score (1423), noticeably below Claude and Gemini for creative and narrative tasks. Output can lean formal and analytical.

Best for: Technical writers, analysts, developers writing documentation, structured report generation.

4. Claude Sonnet 4.6: best mid-tier writing model

Table 5 Download CSV

Arena Creative Writing	1443
Arena Instruction Following	1479
IFEval	89.5
Price	$3/$15 per million tokens
Context	200K tokens

Pros: 80% of Opus writing quality at 20% of the input cost. Strong instruction following (Arena IF: 1479). Non-reasoning architecture, same natural prose style as Opus. Good for teams that want Claude's writing style without the Opus price tag.

Cons: IFEval (89.5) is noticeably below the frontier models. 200K context window is smaller than competitors. Can lose consistency on very long outputs.

Best for: Freelance writers, small content teams, marketing departments with moderate budgets.

5. Grok 4.1: underrated writing contender

Table 6 Download CSV

Arena Creative Writing	1431
Arena Instruction Following	1433
IFEval	93
Price	$3/$15 per million tokens
Context	1M tokens

Pros: Solid IFEval (93) and MMLU (99). 1M context window at $3/$15: the same input price as Claude Sonnet but with 5x the context. GPQA: 97 and MMLU-Pro: 90 give strong factual accuracy.

Cons: Arena scores are middling for writing (CW: 1431, IF: 1433). Less refined prose than Claude or Gemini for creative tasks. Smaller ecosystem and tooling.

Best for: Writers processing large reference documents who want a frontier-capable model at mid-tier pricing.

Use-case breakdown: who should use what

Solo creators (bloggers, newsletter writers, freelancers)

You need one model that handles everything (drafting, editing, repurposing content across formats) and cost matters because you're paying out of pocket.

Recommendation: Gemini 3.1 Pro at $2/$12. Highest creative writing Elo, strong instruction following, and still affordable enough for daily heavy use relative to the premium tier.

Upgrade to Claude Opus 4.6 if writing quality is your primary competitive advantage and you can absorb $375/month at the same volume.

Marketing teams (content marketing, email, social)

You need consistent brand voice across multiple writers, fast turnaround on campaign copy, and the ability to generate many variants for testing.

Recommendation: Claude Sonnet 4.6 for brand-voice work where tone consistency matters. Gemini 3 Flash at $0.50/$3 for high-volume variant generation (A/B test subject lines, social post variants). Route complex strategy docs to Claude Opus 4.6.

Developers (docs, READMEs, technical writing)

You need accurate technical content, proper code formatting, and structured output. Creative flair matters less than precision.

Recommendation: GPT-5.4 at $2.50/$15. Highest IFEval among non-Pro models (96), strong at structured output, and the ChatGPT interface is familiar for developers. For API-generated docs, Gemini 3.1 Pro at $2/$12 is still a strong value option.

How to choose

Need the best possible writing quality: Claude Opus 4.6. Highest instruction following, most natural prose, best editor.

Need great writing on a budget: Gemini 3.1 Pro. Highest creative writing Elo, 12x cheaper than Claude Opus on input.

Need structured or technical writing: GPT-5.4. Highest IFEval (96) among standard-tier models, strong analytical style.

Need a Claude-quality writer at mid-tier pricing: Claude Sonnet 4.6. 80% of Opus quality at $3/$15.

Need high-volume content generation: Gemini 3 Flash. Arena CW: 1461 at $0.50/$3, the best ratio of writing quality to cost.

→ See the full leaderboard · Compare models side by side · Best models by category

---