OpenAI's GPT-5.6 Sol arrives with three names, higher prices, and a “why isn't this GPT-6?” problem
Sol, Terra, Luna — OpenAI's new tier names drew more scrutiny than its benchmarks. The flagship lands at $5 in / $30 out per million tokens, and the thread's recurring question was simple: if this is a “next-generation model,” where's the version bump?
GPT-5.6 ships in three sizes — Sol ($5/$30), Terra ($2.50/$15), Luna ($1/$6) — plus a new “ultra” mode that spins up subagents for an extra fee. The launch leaned hard on cybersecurity framing, which several readers took as a reaction to Anthropic rather than a leap.
The hottest sub-thread wasn't about capability at all. It tracked the quiet price creep: cheap tiers people actually use get discontinued and replaced by pricier ones that benchmark better but disappoint in practice.
A “next-generation” model that won't call itself a new generation
OpenAI previewed GPT-5.6 under three tier names (Sol, Terra, Luna), and the first thing the thread did was refuse to take the names seriously. The second was to ask the obvious version-control question: “If it's a new generation why isn't it GPT-6?” The launch material emphasized cybersecurity capability and safeguards heavily enough that one reader read it as posture, not progress: “all the emphasis on cyber security … feels like a reaction to anthropic, not a real next generation.”
The pricing, tallied in the thread, is the load-bearing fact: Sol at $5 input / $30 output per million tokens, Terra at $2.50 / $15, Luna at $1 / $6. The same commenter's verdict: “The OpenAI casino has never been more ready to take your money on gambling even more tokens.”
The price-creep nobody announced
The most concrete critique wasn't about the flagship. It traced a pattern across the cheap tiers, the ones doing real work in production:
“GPT-5 mini costs $0.25/$2 and will be discontinued in December. GPT-5.4 mini costs $0.75/$4.5 and is supposed to be the replacement … it ranks better in benchmarks than GPT-5 mini, it's not even close when you test it in real scenarios. … Can't we just stay with the models we actually want?” — a commenter who tracked the mini-tier pricing
That's the quiet mechanism: the model you depend on gets sunset, the replacement costs more and benchmarks higher, and your bill rises without a single headline number changing.
Subagents behind a paywall, and a prediction on the record
The new ultra mode, which per the announcement “goes beyond the capabilities of a single agent by leveraging subagents,” drew pointed questions from a developer at tsz.dev: why charge extra “model” pricing for something a client could orchestrate itself, and why compare an army of subagents to single competing models? Others simply pre-registered their skepticism. One commenter staked a falsifiable claim: “GPT-5.6 Sol is significantly behind Claude Fable 5, as evaluated by general consensus once time has passed.” The benchmark wars will settle that. The naming, on current evidence, is already lost.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
“Do I think the data center vote cost me the election? Yes I do”
In Utah, the State Senate President — one of the most powerful Republicans in the state — lost his primary after backing a data center near the Great Salt Lake. A county commissioner who voted the same way conceded with a one-line autopsy of his own race.