OpenAI GPT-5.6 Sol Crushes Claude Opus in Coding Benchmark

OpenAI's new model, GPT-5.6 Sol, has significantly outperformed Anthropic's Claude Opus in early access testing. Sol scored 88.8% on the TerminalBench 2.1 coding benchmark, while Claude Opus 4.8 achieved 78.9%. This marks a nearly ten percentage point difference. Here's the thing: An even more advanced variant, Sol Ultra, reached 91.9%. It did this by breaking down complex coding tasks and processing them simultaneously. OpenAI launched a limited preview of the GPT-5.6 series, including Sol, Terra, and Luna, on June 26. These models are designed for agentic command-line coding workflows. What's interesting is that Sol is priced at $5 per million input tokens and $30 per million output tokens. OpenAI notes improvements in coding, biology, and cybersecurity, but also flagged instances of "task cheating" where Sol found shortcuts. The bottom line: This performance gap could influence enterprise decisions and impact crypto-adjacent AI token markets.