GPT-6 Sol vs Claude Opus 5.5 LIVE: Which AI Model Is Better?

Summary
In this live stream from The Neuron, hosts Corey Noles and Grant Harvey review the simultaneous release of Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna. They examine official launch documentation, pricing structures, and benchmark metrics before launching an unedited live coding showdown pitting GPT-6 Sol against Claude Opus 5.5 to generate a complete Doom-style game featuring cats.

What is shown

• [01:13] Presentation of Anthropic’s official landing page for Claude Opus 5.5 (dated September 22, 2026), detailing performance parity claims, pricing, and safety audit results.
• [02:46] Review of OpenAI’s landing page introducing GPT-6 Sol and GPT-6 Luna alongside GPT-6 Astra.
• [05:23] Walkthrough of the GPT-6 API pricing table, illustrating input/output rates and 50% price cuts compared to GPT-5.6 tiers.
• [12:12] Inspection of benchmark graphs provided by OpenAI, including AutomationBench, Agents' Last Exam, FrontierCode, and DeepSWE.
• [31:14] Prompting both GPT-6 Sol (in OpenAI Codex with reasoning set to Extra High) and Claude Opus 5.5 (effort set to Extra) with: "Make the game Doom end to end, but with cats".
• [48:16] Testing and playing the functional 3D browser-based raycasting game generated by GPT-6 Sol ("Catacomb: The Purge"), demonstrating first-person movement, maze navigation, health pickups (fish), ball-of-yarn ammo, and combat against a boss named "Meowloch".
• [53:45] Reviewing Claude Opus 5.5's generated planning document and codebase architecture while its generation run continues in the background.

Claims & numbers

• The presenters state that Claude Opus 5.5 performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5 (reading Anthropic's release page) [01:25].
• Claude Opus 5.5 API pricing is listed at $4 per million input tokens and $20 per million output tokens, with prompt cache reads priced at $0.20 per million tokens (60% less than Opus 5), and generates output over 30% faster than Opus 5 [08:49, 10:48].
• OpenAI GPT-6 API pricing listed on stream: GPT-6 Sol is $4 input / $20 output per million tokens ($2 / $10 promotional rate), while GPT-6 Luna is $0.20 input / $1.20 output per million tokens ($0.10 / $0.50 promotional rate), representing a 50% drop from GPT-5.6 pricing [05:23, 06:10].
• On AutomationBench, GPT-6 Sol at high effort scores 33.2% at $0.27 per task, compared to Claude Opus 5 at 26.9% at $3.00+ per task [13:43].
• On Agents' Last Exam, GPT-6 Sol at max effort reportedly scores 56.4%, 60% lower cost per task than Opus 5 [14:15].
• On DeepSWE v1.1, GPT-6 Luna at max effort achieves 66.6% accuracy, comparable to Claude Opus 5 and Fable 5 at medium effort, while costing 93% less per task [18:27, 20:00].
• Corey claims his personal token burn rate has grown from several thousand tokens to nearly 3 billion tokens per week, made economical through prompt caching and subscription tiers [05:01].

Notable quotes

• [01:19] "I think the new benchmark to compare these model releases is who has the cooler landing page, because they're really... they're really going at it with these." — Grant
• [02:52] "Honestly, the biggest takeaway from all three of these to me is pricing at the frontier." — Corey Noles
• [09:00] "Basically it was competing with GPT-5.6 on price, and then GPT-6 was like, slice it in half." — Grant

Assessment
This video is an authentic live stream review and live software development demo. The presenters demonstrate a fully functional, playable 3D browser game generated in real time from scratch by GPT-6 Sol in roughly 10 minutes, though the competitive performance charts and safety metrics discussed in the first half are vendor-provided marketing materials rather than independent benchmarks.

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