GPT-6 Sol leaks, screenshots and release rumors

The post jokes that after both companies called for slowing development, they released Opus 5.5 and GPT-6 Sol on the same day, and pokes fun at Grok 4.7.

The same prompt had GPT-6 Sol and Opus 5 each build a robotic hummingbird in three.js; the author notes Sol is cheap with 50% off API pricing, but the comparison is against the older Opus 5.

A long post describes Microsoft's Agent Lightning framework, which intercepts Claude Opus 5.5 and GPT-6 Sol/Luna calls through an API proxy to run reinforcement learning on agent trajectories, lifting Qwen3.5-9B by 14.6 points on SWE-bench.

The author calls Opus 5.5 the current SOTA, says Luna improved while becoming nearly free, places Sol below Astra, and discusses cost per task.

A video shows Opus 5.5, GPT-6 Sol, GPT-6 Luna and GPT-6 Astra together in a single chain reaction machine.

The poster thinks Astra is simply too strong and doubts GPT-6 Sol will be able to replace it, a personal expectation ahead of launch.

Databricks has made OpenAI's GPT-6 Sol and GPT-6 Luna generally available on its platform, citing better cost efficiency on the OfficeQA Pro benchmark.

A tester says GPT-6 Sol is useless for cybersecurity tasks: it wraps up within minutes with superficial findings, while even 5.6 Sol digs deeper.

A user testing GPT-6 Sol with computer use reports that its performance is terrible and far from what was advertised.

GPT-6 Sol scores 97% on the Next.js evals leaderboard, tying Claude Opus 5.5 and Fable 5.1 for the highest success rate, with Opus having the best average cost of the three.

elder_plinius released the full system prompts and tool definitions for GPT-6 Sol Codex, over 294,000 characters of extracted text, with a GitHub link.

AI/ML API posted a video comparing 3D generation from both models, saying Opus 5.5 is clearly better at $4.37 per run versus $0.34 for GPT-6 Sol.