Claude Opus 5 vs GPT-5.6 Sol

On the sticker, GPT-5.6 Sol is the cheaper model: $4 and $20 per million tokens against Claude Opus 5's $5 and $25, about 20% less in both directions. The comparison is less clean than it looks, because per-token prices are only comparable if a token means the same thing on both sides. Anthropic's current tokenizer, introduced with Claude Opus 4.7, fits roughly 555,000 words into a million tokens where the previous one fit about 750,000 - so the same text consumes more tokens on Claude than the headline rate suggests, and the real gap on identical input is wider than 20%. Capacity is near-identical: 1M tokens of context against 1.05M, with both capping output at 128K. Opus 5 is the fresher model, with a reliable knowledge cutoff of May 2026 against Sol's 16 February 2026. Benchmark both on your own prompts and compare cost per completed task rather than cost per token.