How are you supporting the claim that GPT 5.6 is "far more token efficient" than Opus 5? Tokens equal, output is cheaper for Opus 5 ($25/1M) than GPT-5.6-Sol ($30/1M), and it seems to outperform slightly on agentic coding benchmarks.
replyThe first chart in the blog post shows a similar $/performance curve to GPT 5.6.
replyWhere 5.6 has optionality to run much cheaper along the same performance curve at lower thinking levels.
There's a later chart that shows Opus 5 ahead, but seems like an esoteric benchmark rather than for common use. (Novel problem solving)
If they had a more efficient model at coding they would lead with that chart.
Here is one data point for cost:
replyhttps://artificialanalysis.ai/models?cost=intelligence-vs-co...
Here is another data point for output token efficiency:
https://artificialanalysis.ai/models?cost=intelligence-vs-co...
Token cost and token efficiency are two unrelated metrics, and anyways what really matters is neither in isolation - it's cost to complete a task.
replyhttps://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-...
reply
It seems roughly equal according to Anthropic's benchmarks