Is GPT-6 Astra better than Opus 5.5? I checked it on the same tests

Summary
Igor from the Russian-language YouTube channel Студия Игор (Studio Igor) benchmarks OpenAI's GPT-6 Astra across a 6-stage 3D game creation pipeline in Unity and Blender, replicating the exact tests previously run on Claude Opus 5.5 and GPT-6 Sol. He evaluates Astra on 3D modeling, humanoid animation, dinosaur video-reference animation, audio extraction/classification, Three.js level prototyping, and final Unity game assembly. Igor concludes that while GPT-6 Astra produces capable results, Anthropic's Claude Opus 5.5 remains superior overall in quality, cost-efficiency, and execution.

What is shown

• [00:42] Test 1: 3D Vehicle Modeling — Prompting GPT-6 Astra to build a 3D expedition jeep model in Blender from an image reference (jeep-ref.png). Astra builds the chassis, suspension, body, interior, and rolls bars, completing in 31 minutes 30 seconds.
• [01:49] Test 2: Humanoid Character Animation — Taking a Tripo AI character mesh and Mixamo animation pack, then instructing Astra to rig and animate missing clips (entering vehicle, steering left/right) and build an interactive web preview landing page.
• [03:51] Sponsored integration for the Zarub virtual card service.
• [04:51] Test 3: Video-to-Animation (Dinosaur) — Astra analyzes 3 video references (dino-walk.mp4, dino-rear.mp4, dino-attack.mp4), extracts keyframes, and animates a T-Rex 3D model in Blender with walk, roar, attack, and run cycles displayed in an interactive viewer.
• [05:51] Procedural generation in Blender of an environment asset catalog (107 models across 20 asset families, including foliage, ruined walls, and a stone bridge).
• [06:15] Test 4: Audio Library Generation — Astra processes 5 video clips with sound, slices out 32 audio events across 8 categories (footsteps, roaring, stone crumbling, bridge collapse), and displays waveform analysis with anomaly notes in a web UI.
• [07:38] Test 5: Three.js Level Prototype — Astra compiles a graybox prototype using geometric primitives showing a winding road, dinosaur chase sequence, and collapsing bridge.
• [09:01] Final Test: Game Assembly in Unity — Astra attempts full game integration. A single-prompt one-shot generation fails; a second prompt asking for "AAA quality" runs for an additional hour before yielding a playable on-rails driving chase game in Unity 6 with cutscenes, audio, and asset placement.
• [11:38] Account usage dashboard, limit consumption breakdown, and token API cost comparison.

Claims & numbers

• Generation times: GPT-6 Astra generated the 3D jeep model in 31 minutes 30 seconds (~31 min), compared to ~1.5 hours for Claude Opus 5.5 and 34 minutes for GPT-6 Sol (presenter states this is about an hour faster than Opus).
• Environment & Sound outputs: Astra generated 107 models across 20 families for the environment catalog, and extracted 32 sound clips across 8 categories from 5 reference videos.
• One-shot failure: Astra failed to produce an acceptable Unity game in a single prompt; reaching a playable build required a second refinement pass and an additional ~1 hour of processing.
• Account limit usage: Across the entire project on ChatGPT Pro ($200/month plan), Astra consumed 17% of the weekly quota, whereas Claude Opus 5.5 consumed approximately 10% of its weekly quota on its equivalent plan for the same project in the previous video.
• API pricing: The presenter claims GPT-6 Astra's API token pricing is roughly 2.5× more expensive than Claude Opus 5.5.
• Model verdict: The presenter states that despite Astra having the advantage of a second refinement pass, Claude Opus 5.5 remains the best model currently on the market.

Notable quotes

• [02:32] "Вот такой вот лендинг со всеми анимациями подготовила нам GPT-6 Astra." ("This is the kind of landing page with all the animations that GPT-6 Astra prepared for us.")
• [09:07] "Ваншотом сделать эту игру не получилось." ("Making this game in a single shot did not work out.")
• [12:09] "...на сегодняшний день моделька от Anthropic Claude Opus 5.5 — это лучшая модель, которая есть на рынке." ("...as of today, Anthropic's model Claude Opus 5.5 is the best model available on the market.")

Assessment
A hands-on independent review and technical comparison by a game developer evaluating AI models inside Blender, Three.js, and Unity. All generations and software interfaces are shown directly on screen without simulated footage, with honest transparency regarding Astra's initial one-shot failure on the final Unity test.

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