Last Week in AI #336 - Sonnet 4.6, Gemini 3.1 Pro, Anthropic vs Pentagon
Last Week in AI Last Week in AI
Anthropic dropped Claude Sonnet 4.6, and it's landing in the free tier too. Meanwhile Anthropic and the Pentagon are reportedly at odds over how its AI gets used.
Based on reporting by Last Week in AI, Last Week in AI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic just shipped Claude Sonnet 4.6, and the headline isn't just better benchmarks — it's that the company handed a frontier-level coding model to people who aren't paying a dime. Reports point to meaningfully improved coding skills over Sonnet 4.5, the kind of jump that usually gets locked behind a Pro or Max subscription. Instead, free and cheap-tier users get a taste of it too.
That's a notable move given how fast Anthropic has been cycling releases lately. Sonnet 4.5 barely had time to settle before 4.6 showed up, and the company shows no sign of slowing its release cadence. Every few weeks feels like a new checkpoint in the coding-assistant arms race with OpenAI and Google, and Anthropic clearly wants Claude to be the default tool developers reach for, whether they're paying customers or not.
The free-tier generosity is also a pointed contrast to another Anthropic storyline making rounds this week: friction with the Pentagon over how its models can be deployed. Anthropic has built its brand on safety-first positioning, and that stance is now colliding with a defense establishment eager to use frontier AI for military applications. The tension isn't hypothetical — it's a real fault line between commercial AI labs chasing government contracts and the ethical guardrails those same labs promote publicly.
All this is happening alongside chatter about Gemini 3.1 Pro, suggesting Google isn't sitting still either. Taken together, the week reads like a snapshot of where the industry actually is: labs racing to out-ship each other on capability, while simultaneously wrestling with who gets to use that capability and for what. Anthropic giving away strong coding tools for free is a nice consumer win. Whether it can hold the line with the Pentagon is the more interesting question nobody's answered yet.
My take — AI-written commentary, not fact-checked reporting
I run this site because I think open-weight models deserve way more attention than the closed-lab hype cycle gives them, and stories like this prove my point — Anthropic can ship a great free-tier model on Monday and still be quietly negotiating military use cases by Friday. That's not hypocrisy, it's just what happens when your business model depends on both consumer goodwill and government money. Watch the Pentagon deal, not the benchmark chart.”
Read more about this at: Last Week in AI