Anthropic Sonnet 4.6 release: model details and Pentagon feud

Anthropic released Sonnet 4.6 this week, introducing a labeled feature called "deep-thinking tokens" and updates aimed at longer-form reasoning and context handling. Google began rolling out Gemini 3.1 Pro in parallel, marking another commercial push in large model product tiers.

Anthropic framed Sonnet 4.6 as an incremental model update in its Sonnet family, focused on improved handling of extended reasoning chains and token-level strategies that the company is calling deep-thinking tokens. The release coincides with Google making Gemini 3.1 Pro available to more customers, a move that highlights continued vendor competition at the top of the model market.

Dario Amodei, Anthropic's chief executive, addressed public pressure from the Pentagon this week and said the Department of Defense's threats "do not change our position" on AI. Amodei repeated the company stance that safety and deployment policies will guide product decisions, even as regulators and national security actors press for clearer boundaries around military use and procurement.

Product updates

Sonnet 4.6 is presented as a Sonnet-series refinement rather than a ground-up architecture change. Anthropic emphasizes token-level tooling it calls deep-thinking tokens, which it describes as an approach for allocating computation to harder subproblems inside a single prompt. The company highlights gains in multi-step reasoning and context retention for longer documents, while continuing to position Sonnet for consumer and enterprise conversational use.

Google's Gemini 3.1 Pro rollout expands the Pro tier of the Gemini family, with Google focusing on enterprise-grade availability and integrations with workspace tools and cloud infrastructure. Google pitches 3.1 Pro as a stability and capability upgrade for paying users, incorporating latency and throughput enhancements alongside model improvements.

Both vendors are presenting these updates as part of normal commercial iteration: Anthropic with model-level research features and policy positioning, Google with product and enterprise scaling. Customers choosing between Sonnet 4.6 and Gemini 3.1 Pro will weigh performance, ecosystem integrations, pricing, and corporate policy positions when they evaluate adoption.

Why it matters

Anthropic shipping Sonnet 4.6 while publicly resisting Pentagon pressure signals the company will prioritize a combined product and safety narrative rather than ceding decisions to government actors. The near-simultaneous Gemini 3.1 Pro rollout underscores that major cloud and AI vendors continue to compete on both technical increments and customer-facing product tiers. Buyers and regulators will watch how each company balances capability improvements with policy commitments when those capabilities touch defense or national security use cases.