Read time: 4 minutes

Leader, welcome back.

Back to normal scheduling before I travel for the weekend.

Bad news after watching Resident Evil: Claude just found a previously unknown biological system inside viruses. Let's hope San Francisco fares better than Raccoon City!

The story that matters more for your budget: Anthropic and OpenAI both cut prices this week.

On Tuesday, Anthropic launched Claude Opus 5.5.
About 90 minutes later, OpenAI launched GPT-6 Sol and GPT-6 Luna.

Opus 5.5 matches Anthropic's top model, Fable 5.1.
It costs $4/$20 per million input/output tokens.
Anthropic says a typical job costs about 40% less than on Opus 5, because it uses fewer tokens. It also writes more clearly, with less jargon.

Sol costs $2/$10, half the price of Opus 5.5.
Luna costs $0.10/$0.50.
Both are only slightly smarter than the models they replace, and for now they're only in Codex and ChatGPT Work, not the regular ChatGPT app.

Why it matters: both companies are competing on price, because their business customers want to cut AI spend.

Price per token isn't the full picture, though.
YouTuber Nate Herk ran eight real tasks on both. He preferred Opus's output on seven, but Opus took 8h40 and cost $213. Sol took 5h51 and cost about $74. Browser Use found Sol scored higher on its web-browsing test (66.9 vs 59.4) at about 3.5x less cost.

My take: Opus 5.5 is the best all-round model right now. Sol is a step behind.
You don't need to pick one.
Use the strongest model to plan and check work, and a cheaper one for the repetitive steps.

What to measure: cost per completed task. That's model spend plus time, retries, and how often someone had to fix the output.

If you use Claude: Opus 5.5 defaults to medium thinking in the app.
Switch to high for complex analysis.
Claude Code's five-hour usage limits also went up 20%.

The virus discovery

Anthropic says about 950 Claude agents spent 21 hours searching a large DNA database for reverse transcriptases, enzymes that copy RNA into DNA. They then checked what sat next to each one.

They found over 200,000, flagged about 3,500 possible systems, and passed the best 20 to scientists as written reports. One agent spotted an unusual repeating DNA pattern, the same kind of clue that led to CRISPR. Lab tests confirmed it produces small RNAs. Anthropic named it ART. Nobody knows what it does yet.

Why it matters to you: the method transfers.
Humans set the question, AI searched and filtered a huge pile down to a shortlist, and humans checked the shortlist.

You can apply the same approach to a year of supplier contracts ("flag any with auto-renewal or price escalation clauses") or a ledger ("flag transactions that don't match the usual pattern for this supplier").
AI does the first pass.
Your team reviews what it flags.

My take: Dario Amodei says Claude led most of the discovery while humans chose the area and ran the lab tests. Andrew Curran pointed out that lab work still happens at normal speed. The same applies in your business: AI speeds up finding issues, not fixing them.

Most AI spend goes on simple, repeated tasks: coding invoices, categorising expenses, routing queries. A cheaper model often handles these fine.

In June, Lindy's founder Flo Crivello moved all his company's AI work to DeepSeek and saved millions. He basically said you don’t need a genuis to write your email. He later said it didn't work for Lindy's more complex product. Cheap models suit simple tasks, not hard ones.

Here's how to test it.

Then compare. If your current model passes 92 and the cheap one passes 89 at a fifth of the cost, you're paying five times more for three extra correct answers. Multiply by your monthly volume to see the real money.

If those three answers don't matter, switch. If the failures are easy to spot automatically, such as a missing field, send only those back to the expensive model.

No technical team? Send this to whoever manages your AI tools, or ask your software vendor two questions: which model do you use, and what does each task cost? Either way, you'll have a number for your next contract renewal.