Best LLMs for Long-form Content Generation
Sustained compositional skill, voice consistency, coherent extended prose.
Sustained compositional skill, voice consistency, coherent extended prose.
Task-by-task breakdown
Theme Generation
Builds a Ghost-theme colour and font configuration from a logo image plus brand context. Extracts dominant logo colour as primary, picks complementary secondary and contrasting accent, picks Google …
| Model | Quality (% of best) | Confidence | Overpay |
|---|---|---|---|
| MiniMax M3 ★ | 91% | RANKED | best value |
| GPT-5.6 Luna | 94% | HIGH | 2x |
| NVIDIA Nemotron-3 Ultra 550B | 94% | HIGH | 4x |
| GPT-5.6 Terra | 91% | HIGH | 5.5x |
| Claude Sonnet 4.6 | 90% | HIGH | 11x |
| GPT-5.6 Sol | 95% | HIGH | 12x |
| Claude Sonnet 5 best | 100% | RANKED | 14x |
| GPT-5.5 | 93% | HIGH | 21x |
Claim-Referenced Analyst Writing
Pooled TT for analyst-prose synthesis tasks that must preserve [N] workflow-global claim references: cluster_claim_synthesis, chapter_consolidation, topic_report_generation. Wired to the …
| Model | Quality (% of best) | Confidence | Overpay |
|---|---|---|---|
| DeepSeek V4 Pro ★ best | 100% | RANKED | best value |
| GPT-5.6 Terra | 91% | HIGH | 3.2x |
| Claude Sonnet 4.6 | 90% | HIGH | 5.2x |
| Gemini 3.5 Flash | 91% | RANKED | 6.7x |
| GPT-5.6 Sol | 91% | HIGH | 7x |
Substack Newsletter
Pooled TT for Substack opener and summary newsletter generation. Same role/voice; opener is shorter announcement, summary is longer recap with per-article links.
| Model | Quality (% of best) | Confidence | Overpay |
|---|---|---|---|
| NVIDIA Nemotron-3 Nano 30B-A3B ★ | 93% | RANKED | best value |
| MiniMax M3 | 99% | RANKED | 2.9x |
| Tencent Hy3 | 92% | HIGH | 4x |
| DeepSeek V4 Pro | 99% | HIGH | 4.8x |
| GPT-5.6 Luna | 91% | RANKED | 5.1x |
| GPT-5.6 Terra | 90% | HIGH | 9.4x |
| Claude Haiku 4.5 | 98% | RANKED | 12x |
| Qwen 3.6 Plus | 97% | HIGH | 15x |
| Qwen 3.7 Plus | 96% | RANKED | 16x |
| NVIDIA Nemotron-3 Ultra 550B | 98% | RANKED | 17x |
| Claude Sonnet 5 best | 100% | RANKED | 19x |
| Qwen 3.6 Flash | 97% | RANKED | 19x |
| Gemini 3.1 Pro Preview | 96% | HIGH | 19x |
| Gemini 3.5 Flash | 99% | RANKED | 24x |
| Kimi K2.6 | 99% | RANKED | 26x |
| GPT-5.6 Sol | 95% | RANKED | 30x |
| Grok 4.5 | 99% | RANKED | 36x |
| Meta Muse Spark 1.1 | 95% | RANKED | 46x |
| GPT-5.5 | 97% | HIGH | 75x |
Landing Page Section Generation
Designs up to 10 thematic sections for a publication's Ghost CMS landing page. Mixes 1-2 public sections (introductory / overview) and the remainder premium (in-depth analysis). Section names are 2-5 …
| Model | Quality (% of best) | Confidence | Overpay |
|---|---|---|---|
| Tencent Hy3 ★ | 92% | HIGH | best value |
| MiniMax M3 | 97% | MEDIUM | 1.7x |
| Qwen 3.7 Plus | 93% | RANKED | 1.8x |
| GPT-5.6 Luna | 98% | MEDIUM | 2x |
| GPT-5.6 Terra best | 100% | MEDIUM | 3.8x |
| NVIDIA Nemotron-3 Ultra 550B | 94% | RANKED | 4.2x |
| Qwen 3.6 Flash | 90% | RANKED | 5.9x |
| Meta Muse Spark 1.1 | 93% | MEDIUM | 9.3x |
| Grok 4.5 | 95% | RANKED | 9.8x |
Onboarding Chapter Outline Generation
Designs the chapter outline for a new analysis template — 5-10 chapters typically, each with a snake_case code and a detailed user_requirement specifying what to cover, what questions to answer, and …
| Model | Quality (% of best) | Confidence | Overpay |
|---|---|---|---|
| NVIDIA Nemotron-3 Super 120B ★ | 93% | RANKED | best value |
| Tencent Hy3 | 91% | MEDIUM | 1.6x |
| Qwen 3.7 Plus best | 100% | RANKED | 2.9x |
| Qwen 3.6 Flash | 98% | RANKED | 3.3x |
| GPT-5.6 Luna | 95% | RANKED | 4.6x |
| NVIDIA Nemotron-3 Ultra 550B | 99% | RANKED | 6.9x |
| Claude Sonnet 5 | 94% | RANKED | 9.7x |
| GPT-5.6 Terra | 97% | RANKED | 11x |
| Meta Muse Spark 1.1 | 91% | HIGH | 12x |
| Grok 4.5 | 98% | RANKED | 13x |
| GPT-5.6 Sol | 99% | RANKED | 29x |
Author Voice Generation
Crafts a 'soul' document — a detailed personality and voice spec for an AI author persona named after a deceased historical figure. Channels the real figure's intellectual style, values, and …
| Model | Quality (% of best) | Confidence | Overpay |
|---|---|---|---|
| NVIDIA Nemotron-3 Super 120B ★ | 90% | HIGH | best value |
| DeepSeek V4 Flash | 95% | RANKED | 1.8x |
| MiniMax M3 best | 100% | RANKED | 2.1x |
| GPT-5.4 Nano | 90% | RANKED | 4.6x |
| GPT-5.6 Luna | 93% | RANKED | 4.7x |
| NVIDIA Nemotron-3 Ultra 550B | 92% | MEDIUM | 6.7x |
| DeepSeek V4 Pro | 95% | RANKED | 7.6x |
| Qwen 3.7 Plus | 94% | RANKED | 8.2x |
| Claude Sonnet 5 | 94% | RANKED | 12x |
| GPT-5.6 Terra | 92% | RANKED | 12x |
| Gemini 3.5 Flash | 98% | RANKED | 13x |
| Meta Muse Spark 1.1 | 95% | RANKED | 15x |
| Grok 4.5 | 96% | RANKED | 16x |
| Claude Haiku 4.5 | 92% | RANKED | 19x |
| GPT-5.6 Sol | 94% | RANKED | 25x |
| Claude Sonnet 4.6 | 94% | RANKED | 58x |
| GPT-5.5 | 95% | MEDIUM | 117x |
Confidence — how sure we are about the quality score (more judgments + more agreement = higher confidence): RANKED many independent judges scored this model's outputs and their agreement is very high (most confident) — HIGH many judges have scored it and they mostly agree (well-pinned) — MEDIUM enough judges have weighed in to publish, but they disagree more than we'd like (treat with a small grain of salt). LOW-confidence cells are hidden everywhere on the site. See the methodology for the exact thresholds.