Vibe Coding With Claude Opus 5.5 AND GPT 6 Sol

Summary
In this livestream, Matthew Miller from BridgeMind tests Anthropic's newly released Claude Opus 5.5 model across multiple automated vibe-coding and 3D rendering tasks. Midway through the stream, OpenAI unexpectedly releases GPT-6 Sol and GPT-6 Luna, prompting side-by-side prompt evaluations across web games, Blender simulations, and SVG generation.

What is shown

• Benchmark Comparison Table [00:35]: Reviewing initial benchmark results for Claude Opus 5.5 against Fable 5.1, Opus 5, GPT-6 Astra, and GPT-5.6 Sol across CursorBench 4.0, TerminalBench 4.0, FrontendCode v1.1, and GPQA.
• Pricing & Platform Setup [02:35]: Checking Claude Opus 5.5 availability on OpenRouter ($4 / $20 per million tokens) and configuring multiple Claude Code agent sessions in the BridgeMind workspace.
• Automated Remotion Video Generation [29:05, 52:25, 69:40]: Claude Opus 5.5 compiles and renders a programmatic motion graphics promo video using Remotion and generated audio for a BridgeMind merchandise launch.
• 3D Blender Rocket Generation [39:25, 53:50]: Claude Opus 5.5 uses the Blender MCP tool to script and render a 3D SpaceX-style Falcon 9 rocket and launch tower scene.
• Three.js Mario Kart Clone ("Turbo Kart Rally") [42:40, 45:55]: A playable browser-based 3D racing game generated in a one-shot multi-agent prompt with custom vehicles, characters, tracks, and power-ups.
• Breaking Release of GPT-6 Sol and Luna [58:55, 61:55]: Live reaction to the appearance of GPT-6 Sol and GPT-6 Luna in OpenAI Codex and on X.
• Call of Duty Zombies Clone ("Dead Signal") [64:05, 74:00]: A 3D first-person shooter web game generated by Claude Opus 5.5 featuring procedural city streets, weapons, UI, and animated enemy waves.
• OpenAI GPT-6 Model Card & Pricing [79:15]: Reviewing GPT-6 Sol API pricing ($2 input / $10 output per million tokens, 1.05M context window, 128K max output tokens).
• GPT-6 Sol FPS Game ("Dustline Holdout") [91:15]: Running GPT-6 Sol's attempt at the same FPS prompt; the controls fail to register player movement.
• Horror House Game Comparison [101:15, 110:05]: Claude Opus 5.5 generates a fully functional 3D atmospheric exploration horror game ("Horror House") compared against GPT-6 Sol's lower-fidelity attempt.
• BridgeBench Visual Evaluations [134:40 - 138:40, 149:20]: Side-by-side rendering benchmarks (Lava Lamp, Rocket Launch, Sunset Ocean, and Turntable) comparing Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7.
• PS5 Controller SVG Vector Render Comparison [170:40, 176:20, 188:00]: Comparing vector SVGs of a PlayStation 5 DualSense controller generated by Claude Opus 5.5, GPT-6 Sol, and GPT-6 Astra.

Claims & numbers

• The presenter highlights that Claude Opus 5.5 scored 66.4% on TerminalBench 4.0 and 57.8% on CursorBench 4.0 [00:35, 13:20].
• The presenter notes that on CursorBench, Claude Opus 5.5 at medium reasoning effort scores 52.5% ($2.91 per task), beating Claude Fable 5.1 on max effort at 51.8% ($17.28 per task) [20:20, 21:05].
• The presenter states Claude Opus 5.5 API pricing is $4.00 per million input tokens and $20.00 per million output tokens on OpenRouter [02:35].
• According to the Artificial Analysis index shown, Claude Opus 5.5 registers an intelligence score of 58, while GPT-6 Sol scores 48 and Grok 4.7 scores 44 [60:05, 131:05].
• The presenter states that on Artificial Analysis evaluations, Claude Opus 5.5 generates 119,000 output tokens per task [77:40].
• The presenter notes that GPT-6 Sol costs $2.00 per million input tokens and $10.00 per million output tokens (a 50% price reduction compared to GPT-5.6 Sol), with a 1,050,000 token context window and 128,000 max output tokens [79:15].
• The presenter reports GPT-6 Luna costs $0.10 input and $0.50 output per million tokens [79:40].
• The presenter states the BridgeBench rocket launch test cost $1.52 for Claude Opus 5.5 (12m 3s generation time), $0.11 for GPT-6 Sol (1m 12s), $0.29 for Grok 4.7 (11m 22s), and under $0.01 for GPT-6 Luna (56s) [149:05].

Notable quotes

• [21:00] "Opus 5.5 on medium effort is now better than Fable 5.1 on max effort. 52.5% versus 51.8% on CursorBench."
• [59:10] "Double drop confirmed! GPT-6 Sol and GPT-6 Luna just dropped in Codex!"
• [115:50] "Opus 5.5 completely mogs GPT-6 Sol, it's not even a debate."

Assessment
This is a live, unedited multi-hour stream showing real-time coding runs, benchmark scraping, and immediate first impressions of Claude Opus 5.5 and GPT-6 Sol/Luna. The demonstrations are authentic browser and terminal executions using real multi-agent coding harnesses, though the stream format features informal community banter and live debugging.

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