Altman: GPT-5.6 is 54% more token efficient on agentic coding

https://www.cnbc.com/2026/07/09/open-ai-sam-altman-chatgpt-5-6-sol.html

by Topfi • 13 days ago

14 4 13 days ago

Topfi• 13 days ago

Title was shortened and slightly editorialized from "OpenAI’s newest AI model is 54% more token efficient on agentic coding, Altman tells CNBC" for readability.

I cannot watch the interview right now, but the way the sentence is written in the article, it reads equally possible to me that this was in reference to other frontier models on the market or compared to GPT-5.5. The former I'd instantly believe, the latter I am very skeptical of.

I'd be very impressed if they actually managed any reduction as 5.5 is already obscenely token efficient to the point that the reasoning traces seem to compact far less reliably compared to 5.4 as they've become barely comprehensible gibberish.

As more reliable compaction is a stated goal for 5.6, if they solved that and got token usage more efficient on top, that'd make the price to performance equation very much in favor of OpenAI over any other lab. 54% more efficient vs 5.5 would make the pricing for a full benchmark suite run competitive with open-weight models, even if we assume higher per token costs.

nolok• 13 days ago

I agree with you, but it's still a massive point though if they managed to go one step beyong in reasoning while keeping their efficiency, to take the most obvious one the token guzzling behavior of Fable is so big it makes as much headline as its heavy filtering and its actual great capacities when allowed to work.

If OpenAI is saying "oh by the way, upgrading to it won't make your usage massively explode", then I take that as a massive win. I'm on on Claude Max x20 and have accepted that Fable is not for me at its price point and token usage.

Borborygymus• 13 days ago

more efficient than what? I loathe this marketing speak.

paradox242• 12 days ago

This is the routing of parts of a task to cheaper models dressed up as some kind of innovation in efficiency. Why anyone takes anything this guy has to say at face value is beyond me.

Altman: GPT-5.6 is 54% more token efficient on agentic coding

Related Stories