Fiction Eval leaderboard · comparison

Claude Opus 4.8 vs DeepSeek V3.2 for fiction writing

Judged blind on fresh genre scenes (LitRPG, cozy mystery, romance, thriller, xianxia…), July 2026 — three tries per model per scene, passages shown in both orders so position can't sway it, graded by a third AI that isn't either of them.

Head-to-head

Claude Opus 4.8 won the direct matchup 30–0 (with 0 ties) across every prompt and sample.

#ModelScoreRecordSlop
1Claude Opus 4.81927 (likely 1888–1976)150–14–4314.88
6DeepSeek V3.21251 (likely 1201–1302)52–144–1123.24

The score is each model's overall strength across the whole 8-model field (like a chess rating); the range next to it is where the true score most likely sits. Slop is a separate count of AI tells (lower is cleaner) — it isn't part of the ranking.

The same scene, both models

LitRPG — opening of the "dungeon-core scene" scene. Full passages for all models are on the board.

Claude Opus 4.8:

I am three days old and I am nearly empty. [MANA: 2/60] The number sits in the dark behind what I am learning to call my sight. I spent the rest — fifty-eight points, everything — hardening the west corridor, packing loose earth into stone because loose earth frightened me. I do not know why it frightened me. I am new. Many things frighten me. Rain has been falling above for a day. I feel it in the soil the way a tooth feels cold water, a long ache down through my single room and my single corridor and into the pebble of crystal that is, apparently, me. Then the ache changes. [PROXIMITY ALERT — LIVING ENTITY] [Entity has entered the Threshold.] Something drags itself in from the rain. I watch — I have no eyes, but the dungeon is my eyes — as it collapses just inside the corridor I hardened. Small. Bent. Ribs like the tines of a broken rake pushing at grey-green skin. It shivers. Water runs …

DeepSeek V3.2:

Rain seeped through the ancient soil, a constant, dripping pressure against its awareness. The core was a point of warmth and purpose in the dark, a consciousness three days old, defined by a single, driving urge: *harden*. It had spent those days pulling minute specks of mana from the stone, weaving them into a lattice of will along one corridor, five paces long. **Corridor Section Hardened.** **Mana: 0/10.** **Passive Mana Regeneration: 0.1 per hour.** The data was not read, but *felt*, a cold certainty in its being. The core rested, a faint blue gemstone buried in a niche, savoring the solidity of its work. The rain-drip was the only sound, the only change. Then, a new pressure. A weight upon the hardened earth of its entrance. A breach. A living heat, flickering and weak, entered its domain. The core’s awareness snapped to it, a lens focusing. It was small, bipedal, and wretched. Green skin was plastered with mud and rainwater. It shivered violently, ribs showing under a ratty pelt. It …

More

Claude Opus 4.8 full review · DeepSeek V3.2 full review · the July 2026 report · all comparisons

How it works: every pair of models is judged blind on the same scene, with the passages' order flipped so being shown first can't sway it, by GPT-5.4 — a family that isn't on the board, so nobody scores their own side. Each score carries a likely range; overlapping ranges are called a tie. Slop is scored separately by a fixed checklist, not an AI. Full board, every prompt, and the FAQ: thebookfactoryai.com/board. Get each new board by email on the model-drop list.

Writing a book of your own? Book Factory runs the same craft checks on full manuscripts.