GPT-6.1 Sol hands-on: cheap and capable but too slow
The author says Sol handles everyday coding but needs 2-3 iterations for 3D and trails Opus 5.5, runs at only 20-30 TPS, and suggests Astra for architecture planning and Sol for implementation.
🚨 GPT-6.1 SOL — MY TAKE SO FAR . Overall Very capable for the price Handles most everyday coding tasks without much trouble Has the core strengths of Astra, but still feels like a step below it 3D Work Can produce solid 3D results Usually takes 2–3 iterations to get there Opus 5.5 is noticeably stronger here, with Sonnet 5.5 also feeling more consistent The Biggest Problem: Speed 💀 I’m seeing roughly 20–30 tokens/sec Opus 5.5 can push 100+ TPS That gap makes Sol feel much slower than it should Best Codex Workflow Use Astra for architecture, schemas, and planning Hand the actual implementation over to 6.1 Sol The combination works extremely well If OpenAI can seriously improve Sol’s latency, this model could become a much more compelling everyday choice.
This post quotes another post: “GPT-6.1 Sol quick review: 3D needs retries and speed is the weak point”