TabLOID

Anthropic filed its S-1 on October 26 and intends to list on Nasdaq as ANT. Its announcement leaves the offering price and share count undecided; the registration has not yet become effective.

The filing reports $16.3 billion in revenue for the first half of 2026, up 1,300% from a year earlier, alongside a $693 million operating loss. Its much larger $80.7 billion net loss includes $72.4 billion in fair-value changes to convertible notes, derivative liabilities and SAFEs. Those accounting categories should not be confused with operating spending.

Fast growth is the headline. The filing also warns that comparable growth rates may not continue.
Anthropic’s announcement, October 26 · SEC registration statement

GPT-6.1 Terra is available in ChatGPT, Work, Codex and the API. OpenAI lists standard API prices of $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million.

The company says standard API pricing is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. Its own evaluations report improved performance on computer-use tasks. Those are vendor results, not a promise that an agent will finish your particular workflow correctly.

For builders, the useful experiment is a familiar one: give the model a representative job, inspect the result, and count the whole bill, including retries. A cheaper token becomes useful when it helps produce reliable work.
OpenAI’s model announcement and pricing, checked October 26

Canva describes Craft as its first image foundation model trained in-house from scratch. Its process plans a composition before rendering it, with output up to 2K and more than 30 aspect ratios.

The practical pitch is what happens after generation: Magic Layers can separate an image into editable elements. Designers can keep working on the result rather than treating every picture as a sealed box.

Craft is available in Canva and through a public API. Canva labels its MCP connection as coming soon. That distinction matters if you are planning an integration today.

Canva’s performance comparisons are company claims. The more useful test is whether the model can follow your layout, preserve the details that matter and leave you something worth editing.
Canva’s Craft announcement, checked October 26

Block says it trained Dessa from scratch using human speech, without text or a pretrained model. The system can listen and speak at the same time, allowing interruptions and the little sounds that keep a conversation moving.

The company says inference runs locally on a MacBook Pro with an M4 Max chip or better. Its recordings demonstrate changes in timing and expression, but they are selected company examples, not independent benchmarks.

Early access is being offered to a small group of AI safety researchers. Treat Dessa as research to watch, rather than a generally available assistant you can build a business around today.

The interesting question is broader than a convincing voice: how much meaning gets lost when every sigh, pause and laugh must first become a transcript?
Block’s Dessa research page, checked October 26