GPT-5.6 is selling efficiency

OpenAI wants buyers to count completed work

Efficiency is the headline.

OpenAI launched GPT-5.6 in three tiers: Sol, Terra, and Luna, with API prices ranging from $1 to $5 per million input tokens. Its new ultra setting runs four agents in parallel for demanding work, while Sol reached 92.2% on the BrowseComp evaluation.

OpenAI is giving companies clearer ways to match model cost with task difficulty, which should make large deployments easier to budget. The stronger pitch is practical: fewer tool calls, shorter runtimes, and finished files that need less cleanup on a laptop screen.

That changes buying decisions.

Teams may reserve Sol and ultra for complex research or coding, then route routine work through Terra or Luna. Competitors will face more pressure to publish task completion, latency, and cost together.

Will customers pay for peak intelligence, or mainly for fewer failed runs?

Public questions need public answers

Anthropic is asking the public what it should answer.

The company wants people to submit hard questions about AI and says it will publicly track the actions taken in response. It points to surveys of 52,000 Americans and 81,000 Claude users across 159 countries as evidence that the work has already started.

That is useful groundwork. The harder test is whether public input changes product decisions, safety policies, or business priorities after the announcement cycle ends.

A public record could give critics something concrete to check when Anthropic announces layoffs, new safeguards, or more capable models. A dashboard glowing on a laptop matters only when the updates stay specific.

Who gets an answer first?

Meta turned public photos into AI inputs

A public Instagram photo can now become someone else’s AI prompt.

Meta’s new Muse Image tool lets users tag public profiles and generate images using faces found in their posts. Adults with public accounts are included by default, while private accounts and users under 18 are excluded.

That default is the problem. People posted those photos for Instagram, and Meta has quietly expanded the permission into a new product use case.

The setting sits under “sharing and reuse,” where nearly identical toggles can be easy to misread on a phone screen. If this expands to Facebook and video, users may have to keep checking which old posts can feed new tools.

Who should carry that burden?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.