Real head-to-head · same prompt, one shot
Claude Opus 5 vs Qwen 3.8
The new Anthropic flagship — benched on all 45 one-shot builds the day it landed. vs Alibaba's 2.4T flagship — benched through Qoder.
Head-to-head verdict: Claude Opus 5 wins 23–15 with 7 ties.
What I tested — same prompt, two models
I run the same fixed prompt set through every new model the day it drops — same string, one shot, single HTML file out — and I score the result 0–10 on whether it ran, how close it hit the brief, and how good it looked. Below is what came out when I gave the exact same prompts to Claude Opus 5 and Qwen 3.8, side by side, on 45 shared tasks inside the Agent Operating System.
Both models were given identical prompts inside the Agent Operating System — no help, no iteration, no "best of N" tricks. I run each prompt once, save the HTML file the model produces, and score it 0–10 on whether it ran, how close it hit the brief, and how good it looked. The scoring is mine. The verdicts below are pulled from my source comparison guides at agentos.guide where I publish every score and the reasoning behind it.
Claude Opus 5 · Benched on all 45 GoldieBench tasks via API on release day, incremental deploys as scores landed.
Qwen 3.8 · Benched on GoldieBench via the Qoder CLI (`qoder-qwen`, model Qwen3.8-Max-Preview) — the only door while it has no public API. Non-game tasks are one-shot like the rest of the field. GAME tasks run in Qoder's real AGENT mode: skill-infused build, then up to 2 QA fix rounds where a vision judge + live console errors are fed back and Qwen 3.8 edits its own file (it took crypt from a black-screen 3.0 to a torch-lit 7.8). Scored on a mid-play frame by the same Opus judge as everyone else.
Side-by-side on 50 shared tasks
Click any cell to play that model's actual one-shot attempt. Medals are derived from my 0–10 scores per task (highest = 🥇, second = 🥈, third = 🥉).
Task ↓
Claude Opus 5
Qwen 3.8
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Game
Other
Other
Other
Page
Page
Page
Where Claude Opus 5 beat Qwen 3.8
The tasks where I gave Claude Opus 5 a higher 0–10 score on the same prompt — with the actual commentary from my source guides.
Waves
Visual
Claude Opus 5 8.7
·
Qwen 3.8 2.0
(+6.7)
· Gerstner sunset ocean
What I saw: Strong: convincing Gerstner wave surface with sunset foam highlights, floating buoys that bob on the sea, and a polished sea-state control panel with presets. Weak: only 20fps and dense mesh could hurt on low-end devices, but visually it matches/beats the field best.
Game
Game
Claude Opus 5 8.4
·
Qwen 3.8 6.5
(+1.9)
What I saw: Gorgeous low-poly alien arena with a well-modeled mech, polished sci-fi HUD, working minimap showing 5 hostiles, and full combat loop wired in (kills/waves/bolts). Strong and shippable, but the screenshot shows no enemies actively engaged in frame — combat is implied via HUD rath…
Dogfight
Game
Claude Opus 5 8.4
·
Qwen 3.8 6.8
(+1.6)
· polished flight sim
What I saw: Gorgeous low-poly jet, layered HUD, radar with 4 bogeys, gradient sky and terrain with pull-up warning all render cleanly and read as a real dogfight sim; slightly short of top since the screenshot shows enemies only as radar dots rather than visible in-air combat/tracers, and sc…
Nordiccrypt
Game
Claude Opus 5 8.0
·
Qwen 3.8 6.4
(+1.6)
What I saw: Strong first-person crypt crawler with a clear enemy (Crypt Shade with health bar) approaching, first-person axe viewmodel, HUD with health/vigor/kills/minimap, and a torch-lit chandelier — combat is present and shippable. Weak points: the scene reads as dim beige boxes rather th…
Neonsnake
Other
Claude Opus 5 8.3
·
Qwen 3.8 6.8
(+1.5)
What I saw: Strong 3D neon arena with polished glow, additive bloom, HUD, autopilot AI (BFS pathfinding), and clean neon frame/ring aesthetics; snake head and food read clearly with nice lighting. Loses a touch because the snake body barely reads as segmented at score 0 and the floor/border …
Where Qwen 3.8 beat Claude Opus 5
The tasks where I gave Qwen 3.8 a higher 0–10 score on the same prompt — with the actual commentary from my source guides.
Neonblaster
Game
Qwen 3.8 8.7
·
Claude Opus 5 6.8
(+1.9)
· synthwave 3D shooter
What I saw: Stunning outrun-style 3D perspective with pyramids, sun, grid floor and a beautifully rendered ship plus polished HUD, wave banner, and enemies inbound — visually well above the field's typical 2D shooters. Screenshot shows active wave (score 5250, 7 kills, hostiles visible), and…
Arcade
Game
Qwen 3.8 7.8
·
Claude Opus 5 6.5
(+1.3)
What I saw: Ambitious 3D reinterpretation of Snake with genuine combat (visible enemies, threats counter, hull/boost systems) and a highly polished neon HUD; strong presentation but it drifts from the classic arcade brief into a bespoke shooter-serpent hybrid, and the scene reads a bit clutt…
Twilightvale
Game
Qwen 3.8 8.4
·
Claude Opus 5 7.3
(+1.1)
· Atmospheric twilight village
What I saw: Strong twilight atmosphere with cohesive low-poly village, snow, lit lanterns, character model, and polished HUD (minimap, quest, elite bar) all rendering cleanly on-brief. Held just under top tier because the screenshot shows no visible enemies engaged and combat/weather can't b…
Mlx Speedtest
Other
Qwen 3.8 8.4
·
Claude Opus 5 7.4
(+1.0)
· Polished MLX dashboard
What I saw: Strong on-brief render: crisp MLX branding, 6 discovered models with tok/s estimates, phased benchmark steps, a 3D gauge with needle, live log, platform specs and a throughput chart — highly polished and cohesive. Slight weakness is the paused/loading state means the gauge reads …
Matrix
Visual
Qwen 3.8 8.6
·
Claude Opus 5 7.8
(+0.8)
· volumetric depth rain
What I saw: Gorgeous multi-layer depth with blurred far field, white-hot heads, katakana glyphs and an EMP pulse ring — plus polished CRT scanlines, glitch title, themes and full HUD controls. Amber-tinted mutation streaks and rich feature set push it to the top of the field.
Strengths & weaknesses I logged
Claude Opus 5
Strengths
- Frontier-class coding + agentic reasoning (Claude 5 family)
- 1M-token context — reads an entire codebase in one call
- Benched here with skill-infused game prompts the day of release
Trade-offs
- Premium pricing ($5/$25 per M) — route the everyday 90% to cheaper lanes
- Reasoning-by-default eats token budgets unless tuned per call
Qwen 3.8
Strengths
- Skyrim-style open worlds — the Dragon Realm build rendered a lit snowfield, first-person sword and working roaming enemies (7.8)
- Held up across game genres early — voxel sandbox and Doom raycaster both came out shippable
- Runs as a real agentic coder inside Qoder (writes + iterates on files), not just a chat model
Trade-offs
- Torch-lit dungeon (crypt) came out generic (6.3); one open-world RPG one-shot black-screened (twilightvale 2.5) — classic three.js r128 API drift
- Preview is Qoder / Token-Plan only — no OpenRouter or public API, so it can't be routed into an app the way the OpenRouter models can
Pricing & context — the spec sheet
| Spec | Claude Opus 5 | Qwen 3.8 |
|---|---|---|
| Vendor | Anthropic | Alibaba |
| Context window | 1,000,000 tokens | Served through Alibaba's Qoder agent platform; the 3.8-Max preview has no standalone public context window yet. |
| Price | $5 / $25 per M | Qoder plan |
| Pricing detail | Anthropic's brand-new flagship — the first Opus of the Claude 5 family, with a 1M-token context window. Benched via API the day it dropped; game tasks use our skill-infused AAA build prompts. | Qwen3.8-Max-Preview is Alibaba's ~2.4T-parameter flagship, positioned just behind Claude Fable 5. It is NOT on OpenRouter or a public API yet — the only access today is inside Qoder (Alibaba's agentic coding platform, free 2-week Pro trial). Benched here via the Qoder CLI on model `Qwen3.8-Max-Preview`, one-shot. |
| Release | 2026-07 | 2026-07 |
| Bench coverage | 50/50 scored · avg 8.27/10 | 45/45 scored · avg 8.10/10 |
The verdict — which should you pick?
Across 45 scored shared tasks, the averages are essentially tied — Claude Opus 5 8.34 vs Qwen 3.8 8.10. This isn't the comparison where one wins; it's the comparison where you pick based on context, pricing, and what you're actually trying to ship.
If you only run one of these inside your stack, the head-to-head average above is the call. If you can run both, my honest play is to wire Claude Opus 5 and Qwen 3.8 both into the Agent Operating System and dispatch each from the kanban by task type — hardest agentic builds → Claude Opus 5, one-shot 3d game and world prototypes where atmosphere matters → Qwen 3.8. That's the same setup I run for the 4,000+ founders inside the AI Profit Boardroom.
FAQ — Claude Opus 5 vs Qwen 3.8
Which is better, Claude Opus 5 or Qwen 3.8?
On Goldie Bench, Claude Opus 5 averages 8.34/10 across the shared tasks, with 13 gold, 7 silver, 5 bronze overall. Qwen 3.8 averages 8.10/10, with 8 gold, 7 silver, 7 bronze. Claude Opus 5 wins the head-to-head 23–15.
How much does Claude Opus 5 cost vs Qwen 3.8?
Claude Opus 5: Anthropic's brand-new flagship — the first Opus of the Claude 5 family, with a 1M-token context window. Benched via API the day it dropped; game tasks use our skill-infused AAA build prompts. Qwen 3.8: Qwen3.8-Max-Preview is Alibaba's ~2.4T-parameter flagship, positioned just behind Claude Fable 5. It is NOT on OpenRouter or a public API yet — the only access today is inside Qoder (Alibaba's agentic coding platform, free 2-week Pro trial). Benched here via the Qoder CLI on model `Qwen3.8-Max-Preview`, one-shot.
What's the context window for Claude Opus 5 vs Qwen 3.8?
Claude Opus 5 has a 1,000,000 tokens context window. Qwen 3.8 has a Served through Alibaba's Qoder agent platform; the 3.8-Max preview has no standalone public context window yet. context window.
When should I pick Claude Opus 5 over Qwen 3.8?
Pick Claude Opus 5 for: Hardest agentic builds; Whole-repo reasoning; Frontier one-shots. The trade-off is the weaknesses we logged on the bench: Premium pricing ($5/$25 per M) — route the everyday 90% to cheaper lanes; Reasoning-by-default eats token budgets unless tuned per call.
When should I pick Qwen 3.8 over Claude Opus 5?
Pick Qwen 3.8 for: One-shot 3D game and world prototypes where atmosphere matters; Anyone already in the Qoder IDE/CLI wanting a near-frontier model free on the Pro trial; A cheaper stand-in for Fable 5 on creative-visual builds. The trade-off is the weaknesses we logged on the bench: Torch-lit dungeon (crypt) came out generic (6.3); one open-world RPG one-shot black-screened (twilightvale 2.5) — classic three.js r128 API drift; Preview is Qoder / Token-Plan only — no OpenRouter or public API, so it can't be routed into an app the way the OpenRouter models can.
How does Goldie Bench score Claude Opus 5 vs Qwen 3.8?
Every demo on this page was built by Julian Goldie inside the Agent Operating System — same fixed prompt for both models, one shot, single HTML file out. Each result gets a 0–10 score on whether it ran, how close it hit the brief, and how good it looked. The highest score on each task gets gold; second gets silver; third gets bronze. See methodology for full provenance.
Related comparisons
Other head-to-heads using the same scoring system:
Claude Opus 5 vs Fusion Qwen 3.8 vs Fusion Claude Opus 5 vs Hermes MoA Qwen 3.8 vs Hermes MoA Claude Opus 5 vs GPT-5.6 Sol Qwen 3.8 vs GPT-5.6 Sol Claude Opus 5 vs Claude Fable 5 Qwen 3.8 vs Claude Fable 5Full model pages: Claude Opus 5 · Qwen 3.8 · back to the leaderboard
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