OpenAI's new "Ultra" mode arrives — and an employee says the token scoreboards became "use cheaper models" emails
A one-line tweet says GPT-5.6 Sol Ultra is coming to Codex. The interesting reply comes from inside a big company that already has it: the same managers who gamified token spend two months ago now email everyone to spend less.
The announcement is thin — Ultra reportedly uses subagents to go beyond a single agent. The thread fills with access logistics and a guess that Pro-tier already runs ~12 parallel agents synthesized into one answer.
The reversal is a corporate weather report: praise for the biggest token users flipped to weekly emails begging for restraint — arriving just as OpenAI reportedly found a way to halve inference cost.
A thin announcement, one telling reply
OpenAI's Thomas Sottiaux posted that GPT-5.6 Sol Ultra will be available in Codex. Per OpenAI's own preview material cited in the thread, the new "ultra mode" goes beyond a single agent by leveraging subagents to accelerate complex work. Most of the discussion is logistics — individual access, dates — plus a plausible guess about how these tiers work under the hood:
"I think gpt-5.5-pro runs 12x parallel gpt-5.5 agents behind the scene and uses OpenAI's secret sauce to synthesize their answers into one insanely good response." — a commenter who tracks frontier-model behavior
That's the product story. The comment that actually reports news came from inside a company already running Ultra.
The scoreboard that flipped to a spending diet
"2 months ago management was showing us scoreboards, praising leaders who used most tokens. Last few weeks, we're getting weekly emails, telling us that whenever we can - we should use cheaper models" — an employee whose employer already has Ultra provisioned
And a builder connected it to the economics moving underneath the hype:
"I wonder if it's related that that OpenAI has found a way to cut inference costs by half, according to The Information." — a developer who builds LLM systems
What the anecdote actually signals
Put the two together and "Ultra" stops looking like a pure capability flex. A mode that fans work out to many subagents multiplies token consumption — right as enterprises hit the bill and start rationing, and right as the vendor claims a 50% inference-cost cut. The interesting variable for anyone budgeting around these tools isn't the benchmark on the new tier; it's that the same organization can gamify maximal spend and then plead for restraint inside a single quarter. Ultra is the headline. The whiplash on that internal scoreboard is the story.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Those slanted blinking "computer" panels in a hundred movies aren't computers — they're 1950s modems
A database catalogs every real computer that ever appeared on screen, and readers swap favorite sightings. Then John Nagle reveals the running gag hiding in decades of film: the iconic angled panels everyone reads as a supercomputer are the same rented 1950s hardware — and not even the computer part.