OpenAI’s Sol costs half as much as Opus 5.5; Trump renames AI super intelligence

Morning, humans.

Today’s theme: what things cost, and what they’re called.

Anthropic cut Claude Opus 5.5 to $4 per million input tokens on Tuesday. Ninety minutes later, OpenAI priced GPT-6 Sol at $2. Coincidence, surely. Sol also scored lower on the only independent test.

Snorkel AI raised $350 million at a $3.5 billion valuation. Its $375 million run rate takes a recent stretch and assumes the whole year looks like it.

President Trump wants to retire the A in AI. At the U.N., he said U.S. documents will call it “super intelligence.” He polled his followers first. The winning name was not on the ballot.

Stay curious,

Marcus Schuler

Anthropic and OpenAI released Claude Opus 5.5 and GPT-6 Sol 90 minutes apart Tuesday, with Sol priced at half the token rates.

Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. Sol charges $2 and $10 on the same bases. Sensitive cybersecurity requests to Opus 5.5 can route to the older Opus 4.8 without developer visibility.

Artificial Analysis scored Opus 5.5 at 51.2 on its Intelligence Index and $1.34 per task using medium effort. Sol reached 47.5 on the same index and $1.06 per task at maximum effort. The vendors supplied no direct comparison.

Why This Matters:

• API buyers can pay twice the token rate for the higher score, or tune Sol for lower task costs.
• Anthropic plans to add Opus 5.5 to a three-tier Cyber Verification Program. Its fallback rate remains undisclosed.

Snorkel AI raised $350 million in a Series E at a $3.5 billion valuation on Tuesday.

Insight Partners and S32 co-led the round. Snorkel’s September 2026 annualized revenue run rate reached $375 million, more than 18 times its level about a year earlier, though the measure differs from booked revenue and lacks an audit. The shift toward finished datasets and reinforcement-learning environments gives customers a different way to buy expert training material.

Bloomberg follows pension managers trying to diversify when AI exposure sits inside much of their portfolios.

Monte Tarbox of the $327 billion New York City Retirement Systems rejected an AI-heavy private equity fund. Kari Vatanen at Finland’s 36 billion-euro Elo Mutual Pension Insurance said capping AI exposure in the name of diversification risks damaging returns.

• China’s Cyberspace Administration summoned all seven companies named in Anthropic’s September 2026 threat report, then focused on DeepSeek and Moonshot, The Information reported.
• AbbVie signed a multiyear Iambic collaboration on Sept. 21, 2026, for small-molecule drugs across three therapeutic areas, with payment amounts undisclosed.
• OpenAI formed a nine-member mathematics advisory group without decision-making power at the Institute for Advanced Study on Sept. 21, 2026.
• Qualcomm launched Snapdragon 8 Elite Extreme Gen 6 on Sept. 22, 2026, with on-phone support for a 30-billion-parameter mixture-of-experts model.
• DigitalOcean opened Managed Agents in public preview on Sept. 22, 2026, with a $0.060 active-hour example against $0.126 for allocated capacity.
• Hugging Face added native GGUF model loading to Transformers on Sept. 22, 2026, whose Apple Silicon path approached llama.cpp throughput on an M2 Max.

Replace a forecast with action triggers.

A forecast can make uncertainty look settled. Convert it into signals that tell your team when to act.

Your raw input:

The prompt:

Why this works:

The model must connect uncertainty to evidence and a specific response. Thresholds, owners and dates keep the answer useful when conditions change.

What to use:

GPT-5.6 Sol is best for the mixed reasoning and table structure. Claude Opus 5 is a strong fallback for longer source packets.

Gimlet Labs sells an inference cloud that divides an AI workload among different processors for model labs and data-center operators. Its bet is that mixed chip fleets can beat an Nvidia-centered stack on cost and speed, but that advantage rests on company benchmarks.

Founders:

Zain Asgar, Michelle Nguyen, Omid Azizi, Natalie Serrino and James Bartlett founded Gimlet in 2023 after working together at Pixie, the Kubernetes observability startup acquired by New Relic. Asgar had also been an Nvidia GPU architect and Google AI engineering lead.

Product:

Gimlet traces a model, splits inference stages such as prefill and decode, then schedules each piece on suitable hardware. Customers buy Gimlet Cloud or run the managed software in their own data centers. Gimlet says frontier labs and other large inference users pay for it, but names none.

Financing:

Gimlet had raised $392 million by September 4, 2026. Its latest financing was a $300 million Series B led by Andreessen Horowitz on that date, at a company-stated $3 billion valuation. Gimlet says it added “billions” of dollars in contracted revenue between March and September 4, 2026, without disclosing contract values or duration.

The risk:

Gimlet must beat Nvidia’s integrated software stack and rival inference clouds while building mixed-chip data centers. Its September 4 claim of 5-to-10-fold speedups at the same power footprint remains company-reported, leaving delivery economics untested in public.

Credit: Civitai

Prompt (excerpt): a young woman in a heavy, worn olive-green parka holds a rifle in a desolate, foggy field of dry grass, beside a large, rusty humanoid robot covered in mud and wear.

President Trump told the U.N. General Assembly on Sept. 22, 2026, that the U.S. government will call artificial intelligence “super intelligence.”

Our take: The White House asked followers to pick a new name for AI. The ballot offered superior intelligence, extreme intelligence and supreme intelligence. The President then chose super intelligence, which was not on it. It spares everyone the counting.

The new name has a second problem. Superintelligence already names AI that outthinks humans. Several frontier labs say they are building it, and AI safety researchers call it their chief worry. In the same speech, Trump said the U.S. “totally rejects any attempt to construct a globalist scheme to control” it. So Washington adopted the safety camp’s scariest word and ruled out a global leash before lunch. Nobody polled the safety camp.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai