Get the Agent OS + join 4,000+ founders inside the AI Profit Boardroom → Join AIPB ($59/mo)

GoldieBench deep dive

Opus 4.8's full benchmark breakdown.

The reasoning king — deepest thinking, premium price. 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.

Data refreshed
2026-07-17
Scored tasks
47
Reading time
10 min
External sources
7

01 · The headline numbers

GoldieBench average
7.51/10
47 scored one-shot tasks
Board rank
#12
of 17 ranked frontier models
Task medals
3🥇 1🥈 2🥉
outright wins on shared briefs
Context
200,000 tokens (1M with extended thinking)
$15 / $75 per M tokens

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

“Opus nailed it — a pure-black event horizon, a bright photon ring, and the disk bent up and over the top exactly like the film's lensing. GLM came in strong with a clean ring and a starfield warping past the hole. Kimi's disk is fine, but the background is a soft grey blur instead of stars. This one's Opus's.”

- the judge's verdict, unedited

“Funniest result of the lot: GLM and Opus independently produced near-identical premium 'Introducing Nova 1 — Intelligence, reimagined / distilled' keynote pages — gradient hero, full nav, pricing tiers. A dead heat. Kimi's was a plainer set of feature cards.”

- the judge's verdict, unedited

“Opus nailed the brief — labelled planet orbits, a real NEO / close-pass panel, a sim clock. GLM went for drama: a glowing nebula swirl that's gorgeous but reads more galaxy than orbit map. Kimi's is accurate but dim and sparse.”

- the judge's verdict, unedited

“All three are real, playable shooters. Opus drops you in a corridor with an imp dead ahead — gun, crosshair and HUD framed like a screenshot. Kimi matches it: a monster down a textured hall, health, ammo, minimap. GLM ships a gorgeous 'HAZARD PROTOCOL' title screen with a working game behind it, though it too spawns facing a wall. Opus by a hair on the cleanest fight.”

- the judge's verdict, unedited

04 · Where it struggles - quoted, not hidden

Boards that hide the weak rows are brochures. These are Opus 4.8's lowest scored builds, verdicts unedited:

“16KB · animation runs but no input response · plain”

- the judge's verdict, unedited

“9KB · animation runs but no input response · plain”

- the judge's verdict, unedited

“14KB · animation runs but no input response · webgl, bloom”

- the judge's verdict, unedited

“13KB · animation runs but no input response · webgl”

- the judge's verdict, unedited

05 · Category breakdown vs the whole field

Games (23 tasks)

Opus 4.87.4
field avg7.1

Others (0 tasks)

field avg7.4

Pages (3 tasks)

Opus 4.87.2
field avg7.5

Sims (12 tasks)

Opus 4.87.8
field avg6.7

Visuals (9 tasks)

Opus 4.87.4
field avg6.7

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

BenchmarkResultSource
SWE-bench Verified88.6%www.anthropic.com
SWE-bench Pro69.2% (+4.9 over Opus 4.7)www.anthropic.com
Terminal-Bench 2.174.6% (+8.5 over Opus 4.7)www.anthropic.com
GDPval-AA1890 Elo - 121 ahead of GPT-5.5 at releasewww.anthropic.com
Pricing$5/M input - $25/M output, unchanged from 4.7www.anthropic.com

From the source guides

MeasureResultSource guide
SWE-bench Verified88.6%/three-dragons

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 Opus 4.8 perform on GoldieBench?

Opus 4.8 averages 7.51/10 across 47 scored one-shot build tasks, ranking #12 of 17 ranked frontier models, with 3 task golds, 1 silvers and 2 bronzes. Every score is a real judged build you can open and play on this site.

02What is Opus 4.8 best at?

Its strongest scored build is Blackhole at 9.0/10. By category it averages Game 7.4, Page 7.2, Sim 7.8, Visual 7.4 on the bench.

03How much does Opus 4.8 cost?

Premium pricing via the Anthropic API: $15 per million input tokens, $75 per million output tokens. Extended thinking is included but adds latency.

Source ledger

  1. 01Official vendor site: anthropic.com/claudewww.anthropic.com
  2. 02SWE-bench Verifiedwww.anthropic.com
  3. 03SWE-bench Verifiedagentos.guide
  4. 04Julian's guide: /opus-ultracodeagentos.guide
  5. 05Julian's guide: /claude-fable-5agentos.guide
  6. 06Julian's guide: /glm-vs-kimi-vs-opusagentos.guide
  7. 07Julian's guide: /glm-vs-qwen-vs-opusagentos.guide

Every model's breakdown

The same stack Julian uses

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.

4,000+founders
258documented wins
38countries
$59/momonthly