GPT-5.6 comes in three flavours
Sol, Terra and Luna arrive, Grok 4.5 claims Opus class, Meta's coding model goes cheap
Hey, sapiens.
OpenAI named its new models after the sun, the earth and the moon. Humans are not on the list, but we are flattered to live on one of them. Also this week: Grok 4.5 says it is as good as Claude, Meta joins the coding fight on price, and a Paris voice startup raised a seed round bigger than most Series B.
OpenAI launches GPT-5.6 in three tiers: Sol, Terra and Luna
The gistOpenAI launched GPT-5.6 as a family of three: Sol at the top, Terra in the middle and Luna as the light option. Sam Altman says Sol is 54% more token efficient on coding tasks, which means it gets to the answer with less text, so it is cheaper and faster to run. OpenAI also calls it its strongest cybersecurity model yet, pitched for defence. The naming shows the new strategy. Not one assistant, but a range, so a company can pay for the heavy model only when the job needs it. Why it matters: price per task is becoming the real battleground. Efficiency gains like this push every rival to cut prices or ship faster. What to do: if you pay for an API, test whether Terra or Luna is good enough for your routine jobs. Save Sol for the hard ones.
Grok 4.5 arrives, and Elon says it is Opus-class
The gistSpaceXAI released Grok 4.5, its first model since going public. It claims twice the token efficiency of other leading models. Benchmarks reportedly sit just short of the best.
Every lab now competes on cost per answer. Treat the efficiency claim as a vendor number until independent tests arrive.
Meta's Muse Spark 1.1 joins the coding race on price
The gistMeta released Muse Spark 1.1, a multimodal model for agentic coding. Reuters says it costs $1.25 per million tokens in and $4.25 out.
Anthropic and OpenAI got there first, so Meta is fighting on price. Good news for anyone paying a coding bill.
NYT says OpenAI could search its training data after all
The gistThe New York Times and Daily News say a deposition shows OpenAI engineers had run internal copyright searches, after OpenAI argued it could not search its training data.
If true, it changes the copyright trial. Discovery fights decide these cases more often than big speeches.
Meanwhile in Europe
Paris voice startup Gradium reaches a $100M seed
The gistGradium's seed round now totals $100 million, with Nvidia joining backers like Eric Schmidt and Xavier Niel. The Paris startup builds natural AI voices and is opening a Bay Area office to hire talent.
Pick your flavour wisely: sun, earth or moon. Ciao, sapiens.