AI spend at Sun Valley has emerged as the defining conversation of this year’s Allen & Co. conference, with OpenAI CEO Sam Altman telling CNBC’s Julia Boorstin on ‘Squawk on the Street’ on 9 July that cost and return on investment have become the dominant questions among the billionaires and executives gathered in Sun Valley, Idaho.
Altman said this was the first year AI spend had been a ‘very big topic’ at the annual summit. ‘Everyone’s asking what we can do to help reduce spend or increase value,’ he said, adding that OpenAI is seeing enterprises begin to prioritise efficiency and demonstrable returns from their AI investments.
AI Spend at Sun Valley and the GPT-5.6 Design Brief
Altman said cost pressure was a direct input into building the GPT-5.6 model family, which OpenAI released publicly on 9 July. The family comprises three tiers: Sol, the flagship; Terra, described as a balanced everyday option; and Luna, the cost-efficient tier.
According to Windows Forum, a limited preview of GPT-5.6 had begun on 26 June, restricted to a small group of partners with what OpenAI characterised as ‘advance engagement and coordinated testing, not a formal federal approval process.’ General availability followed on 9 July.
The naming architecture itself is deliberate. Space Daily reports that OpenAI explained: ‘the number identifies a model’s generation, while Sol, Terra, and Luna identify durable capability tiers that can advance on their own cadence.’ The company also released an agent called ChatGPT Work alongside the model family, designed to carry out whole jobs rather than merely answer questions.
On the flagship tier, Altman told CNBC that GPT-5.6 Sol is ‘not only the best model in the world for most people’ and ‘as good or better as the other best models out there.’ He said Sol is 54% more token efficient on agentic coding tasks, though he did not specify what the comparison baseline was.
The public launch also ended an earlier period of restricted access. CNBC reported that the GPT-5.6 series had initially been limited to a small group of trusted partners whose participation had been shared with the government before the 9 July opening.
Every Enterprise Is Watching the Numbers
Altman’s Sun Valley remarks reflect a broader shift in how large organisations are approaching AI procurement. He said ‘every enterprise’ is now thinking about ‘spend and the value they’re getting in exchange for AI.’
That shift is playing out in public statements from executives across the industry. Coinbase CEO Brian Armstrong said in June he was experimenting with cheaper Chinese models as defaults for his engineers, routing prompts to whichever model was most appropriate to avoid burning unnecessary AI tokens.
Vercel CEO Guillermo Rauch told TechCrunch in an interview released earlier this week that companies need to partner with multiple AI labs for different parts of their stack, drawing on OpenAI, Anthropic, Gemini, and Chinese providers to extract the best value at each layer.
The Sun Valley conference, which opened on Tuesday, brought together some of the technology industry’s most prominent figures, including Netflix co-CEO Ted Sarandos, Uber CEO Dara Khosrowshahi, Apple CEO Tim Cook, and Amazon CEO Andy Jassy.
The GPT-5.6 benchmarks cited by aggregator CometAPI, drawing on OpenAI’s own data, give a fuller picture of where Sol sits competitively, though those figures should be read alongside OpenAI’s primary documentation.
With enterprise procurement teams now scrutinising token costs and model routing, how OpenAI prices the Luna tier against rival offerings may prove the sharper commercial test than any benchmark Sol can clear.