GoldieBench deep dive
Claude Fable 5's full benchmark breakdown.
The newest Anthropic model — first Mythos-class made generally available. Every number below is either one of my own judged one-shot builds - playable on this site - or an outside result with its source linked. Nothing display-only, nothing vibes.
01 · The headline numbers
02 · Every benchmark, bar by bar
All 47 scored tasks, best first. Hover for the judge's comment; click through for every model's build on that task. Or overlay other models in the interactive graphs.
03 · Where it wins - with the judge's own words
“AAA rebuild is showcase-grade: a dramatic Nordic dusk valley with layered snow-capped peaks, pine forests and ruins, a first-person sword AND studded wooden shield in real PBR wood/metal/leather, drifting clouds and snowfall, and a cohesive HUD (compass strip, objective card, HP/stamina/magicka bars). Walk and look verified. Only the absent on-screen wolves keep it from a flawless read.”
“AAA rebuild is an authentic textured Minecraft: procedurally-textured grass/dirt/stone/wood blocks (not flat colors), a blocky tree, flowers and stone structures, a first-person hand with a placement crosshair, a day clock, and a full HUD with a heart row, a hunger row and a textured 5-slot block hotbar. Break, place and move verified — a real little world.”
“AAA rebuild via the threejs-game-director skill transforms it into a real torch-lit crawler: warm-lit cracked-stone walls, a stone sarcophagus, a flaming torch held in first-person, scattered rubble and soul-gem pickups, drifting dust, and a cohesive HUD (vitality, objective + distance, compass, gem counter). WASD/mouse-look verified. Needed a one-line patch to a glitched emissive token and the interior reads a touch boxy, keeping it just under the very top.”
“Gorgeous synthwave palette with banded retro sun, parallax mountains, glowing palms and neon-edged pseudo-3D road with a curving horizon and traffic — genuinely polished and on-brief; only nit is the player car sits a touch high with a slightly floaty tail-light bar, but this clearly matches or beats the field best.”
“Textbook Cornell box with convincing colored-wall bleeding, a golden metal sphere with sharp reflections showing the room, and a glass sphere with proper refraction/caustics and a purple sphere visible through it — physically-correct GI, soft shadows and checker floor all render cleanly. Minor Monte-Carlo noise at 80 samples is expected and it converges; a top-tier build on this task.”
04 · Where it struggles - quoted, not hidden
Boards that hide the weak rows are brochures. These are Claude Fable 5's lowest scored builds, verdicts unedited:
“The tunnel core with depth-fading rings and colorful particles reads convincingly as a wormhole and the UI/title are clean, but the flat teal fog fills most of the frame instead of an immersive tunnel, and the oversized foreground particle squares plus the misplaced solid-tube geometry look like rendering artifacts that break the polish.”
“Renders with a slick animated 3D starfield, floating shapes, and clean glassmorphic feature cards, but the screenshot captures the features section with the torus knot visually intruding over card text — a composition/legibility issue that undercuts polish. Strong tech and theme cohesion, but the overlapping 3D object bleeding into content keeps it below the field's best.”
“Clean starfield with subtle gradient and a nicely styled title, plus solid physics/palette variety in code (rings, sparks, flash), but the captured moment shows an empty sky with just one faint burst starting in the corner, so the screenshot lacks the wow-factor of a mid-explosion capture. Competent and shippable but not visually spectacular enough to top the field.”
“Renders cleanly with a polished neon HUD (speed/score/rings/fury), a large glowing ring, floating decor and a green winged dragon — all brief elements present. Weak spot is the camera framing: the dragon is oddly close and awkwardly angled with wings dominating the frame, giving a less-controlled, generic feel versus the field's best.”
05 · Category breakdown vs the whole field
Games (23 tasks)
Others (0 tasks)
Pages (3 tasks)
Sims (12 tasks)
Visuals (9 tasks)
06 · Outside signals - every row sourced
My bench measures one thing: judged one-shot builds. These outside rows measure other things - kept separate, never blended into the GoldieBench average, every value linked to where it comes from.
Vendor-published numbers
| Benchmark | Result | Source |
|---|---|---|
| SWE-bench Verified | 95.5% | www.anthropic.com |
| SWE-bench Pro | 80.3% | www.anthropic.com |
| Terminal-Bench 2.1 | 88.0% | www.anthropic.com |
| FrontierCode Diamond | 29.3% | www.anthropic.com |
| Humanity's Last Exam | 59.0% no tools - 64.5% with tools | www.anthropic.com |
| GDPval-AA | 1932 Elo | www.anthropic.com |
Independent evaluations
| Benchmark | Result | Source |
|---|---|---|
| Vendor-reported caveat | the 80.3% SWE-bench Pro figure is Anthropic's own - no independent evaluator has published a Fable 5 score on it yet | techjacksolutions.com |
| LMArena Frontend Code Arena | displaced to #2 by Kimi K3's debut; #1 in Gaming domain | x.com |
From the source guides
| Measure | Result | Source guide |
|---|---|---|
| One-shot pass, before agentic repair of 8 crashed/weak tasks | 7.72 avg | /fable-5-plus |
| SWE-bench Verified | 95.0% | /three-dragons |
| Plan quality (Kilo rubric) | 9.1/10 | /kilo-split |
| DRACO deep-research (solo) | 65.3% | /fusion-api |
07 · Head-to-head records
08 · Methodology + honest limits
Every GoldieBench score is a one-shot, single-file build from an identical prompt - no retries, no hand-fixing - rendered for real, screenshotted, and scored 0-10 by one judge model on one rubric across the whole field. Failures score as failures. What this bench does NOT measure: multi-turn agent work, long-context recall, or API latency - that is what the sourced outside rows are for. Full method: /methodology.
Frequently asked questions
01How does Claude Fable 5 perform on GoldieBench?
Claude Fable 5 averages 8.10/10 across 47 scored one-shot build tasks, ranking #5 of 17 ranked frontier models, with 4 task golds, 2 silvers and 2 bronzes. Every score is a real judged build you can open and play on this site.
02What is Claude Fable 5 best at?
Its strongest scored build is Skyrim at 9.0/10. By category it averages Game 8.2, Page 7.6, Sim 8.1, Visual 8.1 on the bench.
03How much does Claude Fable 5 cost?
Released alongside Mythos 5 on June 9, 2026 as the publicly-available member of the new Mythos class. Premium per-token pricing on the Anthropic API; available everywhere Opus 4.8 ships.
Source ledger
- 01Official vendor site: anthropic.com/claudewww.anthropic.com
- 02SWE-bench Verifiedwww.anthropic.com
- 03Vendor-reported caveattechjacksolutions.com
- 04LMArena Frontend Code Arenax.com
- 05One-shot pass, before agentic repair of 8 crashed/weak tasksagentos.guide
- 06SWE-bench Verifiedagentos.guide
- 07Plan quality (Kilo rubric)agentos.guide
- 08DRACO deep-research (solo)agentos.guide
- 09Julian's guide: /claude-fable-5agentos.guide
- 10Julian's guide: /claude-fable-5-use-casesagentos.guide
- 11Julian's guide: /claude-5-agent-osagentos.guide
- 12Julian's guide: /claude-5-ai-seoagentos.guide
- 13Julian's guide: /claude-hermes-olympus-protocolagentos.guide
Every model's breakdown
Run this stack yourself.
Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 4,000+ founders shipping with it every day all live inside the AI Profit Boardroom.