NEW Opus 5.5 vs GPT-6 Astra Building Video Games (NOT Close)

Summary
In this comparative review, presenter Brendan Jowett benchmarks Anthropic's Claude Opus 5.5 against OpenAI's GPT-6 Astra across five increasingly complex video game development tasks generated from identical single prompts. Both models were tasked with generating all C++ code and creating all 3D assets natively in Blender without external downloads or human code intervention. Jowett tests and plays each generated game side-by-side, analyzing build times, API costs, code volume, graphical fidelity, and gameplay mechanics.

---

What is shown

• Rules and Methodology [00:27]: Both models were run on maximum reasoning settings (Claude Opus 5.5 on "Max Effort" and GPT-6 Astra on "Astra Ultra Mode"), generating all 3D assets in Blender, using zero external downloads and zero human-written code.
• Round 1: Rampart (2D Castle Platformer) [00:46]:
  • Comparative stats dashboard displayed at [00:47].
  • GPT-6 Astra gameplay [01:06]: Functional 2D platformer with double jumping, stomping enemies, and basic UI.
  • Claude Opus 5.5 gameplay [02:49]: Polished retro-style graphics, complex UI branding, fluid animations, and custom sound design.
• Round 2: Apex Circuit (3D Arcade Coastal Racer) [04:40]:
  • Comparative stats dashboard displayed at [04:41].
  • GPT-6 Astra gameplay [04:53]: Functional 3D arcade racer with AI opponents, drift/boost mechanics, but simpler low-poly trees and environmental textures.
  • Claude Opus 5.5 gameplay [06:05]: Rich lighting, motion blur effects, customized UI tachometer, detailed sports car models, and competitive AI pathing.
• Round 3: Void Wing (6-Axis Space Dogfight) [08:24]:
  • Comparative stats dashboard displayed at [08:25].
  • GPT-6 Astra gameplay [08:37]: Space combat around a gas giant targeting turrets and enemy fighters amid asteroid belts.
  • Claude Opus 5.5 gameplay [09:40]: Cinematic space dogfight with volumetric nebulae, detailed ship models, dynamic lighting, lock-on targeting mechanics, and explosive debris.
• Round 4: Dead Signal (First-Person Survival Shooter) [11:14]:
  • Comparative stats dashboard displayed at [11:15].
  • GPT-6 Astra gameplay [11:18]: Defending a radio tower from drone spiders in a snowy outpost, featuring basic enemy AI pathing issues [11:50].
  • Claude Opus 5.5 gameplay [12:55]: Stylized survival horror environment with dynamic flashlight illumination, siren sound design, multiple weapons, and aggressive drone swarm AI.
• Round 5: Colossus (Third-Person Golem Boss Fight) [15:10]:
  • Comparative stats dashboard displayed at [15:11].
  • GPT-6 Astra gameplay [15:18]: "Aurion, The Last Colossus" assembly animation, telegraphing shockwave circles and sword attacks.
  • Claude Opus 5.5 gameplay [16:21]: "Kharos, The Stormbound Colossus" cinematic lightning intro, destructible arena floor, attack animations, and glowing weak-point hit mechanics.

---

Claims & numbers

• Anthropic's Claude Opus 5.5 announcement post is cited stating it performs at the level of Claude Fable 5.1 on most tasks and costs 40% less to run than Opus 5 [00:03].
• Game 1 ("Rampart"):
  • Claude Opus 5.5: 70.8 min total build time, 45.9 min to first playable, $48.95 API cost, 4,039 lines of C++ [00:47].
  • GPT-6 Astra: 33.4 min total build time, 18.2 min to first playable, $23.02 API cost, 895 lines of C++ [00:47].
• Game 2 ("Apex Circuit"):
  • Claude Opus 5.5: 117.6 min total build time, 57.7 min to first playable, $67.30 API cost, 5,289 lines of C++ [04:41].
  • GPT-6 Astra: 72.9 min total build time, 16.8 min to first playable, $63.08 API cost, 2,029 lines of C++ [04:41].
• Game 3 ("Void Wing"):
  • Claude Opus 5.5: 99.3 min total build time, 48.3 min to first playable, $55.65 API cost, 5,302 lines of C++ [08:25].
  • GPT-6 Astra: 39.1 min total build time, 18.9 min to first playable, $28.05 API cost, 1,556 lines of C++ [08:25].
• Game 4 ("Dead Signal"):
  • Claude Opus 5.5: 75.0 min total build time, 60.2 min to first playable, $49.88 API cost, 5,050 lines of C++ [11:15].
  • GPT-6 Astra: 42.6 min total build time, 22.3 min to first playable, $32.98 API cost, 1,297 lines of C++ [11:15].
• Game 5 ("Colossus"):
  • Claude Opus 5.5: 94.6 min total build time, 47.0 min to first playable, $47.00 API cost, 5,253 lines of C++ [15:11].
  • GPT-6 Astra: 44.9 min total build time, 22.4 min to first playable, $28.69 API cost, 1,824 lines of C++ [15:11].
• The presenter notes that while GPT-6 Astra built games faster and at lower API costs in most tests, Claude Opus 5.5 consistently wrote roughly 2.5× to 4× more lines of C++ code, producing significantly higher visual and mechanical complexity.

---

Notable quotes

• [00:43] "And spoiler, the results I got were not close at all."
• [06:28] "Once again, a really insane difference. It's like not even comparable, these models, which is really surprising because Astra is also really good at creating these games, but Opus has honestly just come in and really crushed it..."
• [13:19] "This looks like a legitimate game... If I bought this, I would not be thinking at all that this was made by AI."

---

Assessment
This is an independent creator benchmark and review demonstrating end-to-end autonomous game generation by two frontier reasoning models. The builds are shown live and fully playable with detailed telemetry and API pricing metrics, though gameplay was limited to brief playtests of single-prompt outputs.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.