In a nutshell: Anthropic launches Claude Sonnet 5 as an agentic-optimized standard mid-tier with 1M context and promotional pricing, promising previous high-end features at Sonnet cost, but showing benchmark weaknesses in tokenizer efficiency.
Anthropic has launched Claude Sonnet 5 as the new standard mid-tier and immediately rolled it out across Claude, Claude Code, API and partner systems. The model comes with one million token context, new pricing starting at $2/$10 (promo until 31.8./1.9.) and promises agentic capabilities such as autonomous execution and tool use that previously required larger models.
Sonnet 5 is positioned by Anthropic as “the most agentic-capable Sonnet yet” and is meant to offer capabilities such as planning functions, browser and terminal tool use, and autonomous execution that previously required larger and more expensive models. The model is immediately available in Claude Code for Pro users and on the Claude platform including API and Managed Agents.
The pricing model sets Sonnet 5 at $3 per million input tokens and $15 per million output tokens unchanged. In parallel, Anthropic is offering a promotion at $2/$10, valid until August 31 or September 1 (depending on source) – this discount applies directly to API costs. The model features a 1-million-token context window.
The rollout chronology shows typical leak and pre-release patterns: Sonnet 5 initially surfaced through code sightings and leaks, with knowledge cutoff specifications of January 2026, ahead of the official launch. The model became visible in client selectors, Claude Code 2.1.197, Anthropic GitHub and later in production accounts in Germany and worldwide.
In parallel with the model release, Anthropic has expanded the platform: Claude Desktop now supports Linux (Ubuntu/Debian Beta) and brings Claude Code, Cowork and Chat for paid plans – Computer Use is not yet included in this Linux version. Managed Agents also received updates with streaming session deltas, per-session overrides, webhook events and new observability.
Among CTOs, a discrepancy between technical positioning and real-world performance has already been noted: benchmark comparisons point to efficiency issues, including tokenizer changes and 3–6x longer interaction turns in some scenarios. The Fable 5 topic, which had previously triggered intense speculation about a more heavily regulated or regionally gated model, was not announced at launch time – only that Fable/Mythos 5 had been reapproved for release following consultation with authorities.
Source: www.latent.space · Published July 1, 2026
Lumi AI News — AI-assisted curation in accordance with Art. 50 EU AI Act. Paraphrase and classification via Lumi News Pipeline v1.7.2.