Sonnet 5’s Promotional Pricing Ends August 31
The clock is running on Claude Sonnet 5’s introductory rates. Anthropic confirmed the promotional $2 / $10 per million tokens pricing ends August 31, 2026, with standard $3 / $15 rates taking effect September 1. On list price alone, that’s a 50% increase on both input and output.
The catch is subtler. Sonnet 5’s updated tokenizer maps the same content to 1.0–1.35× more tokens than earlier versions, so the effective cost of an identical workload can climb more than the headline change implies. Teams running Sonnet 5 in production should re-benchmark their real token counts now, before the September bill lands — budgeting off the old numbers could leave a nasty surprise.
UST Joins Cognizant as a Global Premier Claude Partner
The enterprise land grab continues. Anthropic named UST as its second Global Premier Partner, the top tier of the Claude Partner Network, with a mandate to move enterprise AI projects from proof-of-concept to production scale. It arrives just days after Cognizant took the same tier — two systems integrators now anchoring Anthropic’s enterprise distribution.
The customer roster kept growing too, with Allianz added to Anthropic’s list of enterprise wins. The pattern is consistent: rather than sell seat-by-seat, Anthropic is wiring Claude into the integrators and platforms that large, regulated companies already buy through — the unglamorous distribution work that quietly decides who wins enterprise AI.
Rate Limits Rise, Tiers Consolidate, and API Keys Get Expiration Dates
Anthropic raised API rate limits across the board. Claude Sonnet and Haiku limits now match Opus at every usage tier, and the old ladder of tiers has been consolidated into three: Start, Build, and Scale. Fewer arbitrary ceilings to trip over as workloads grow, and a simpler mental model for planning capacity.
On the security side, the Claude Console added API key expiration settings: admins can now set an expiration when creating a standard or Admin API key — preset windows, a custom duration, or Never. It’s a small governance win that closes a long-standing gap for teams that rotate credentials on a schedule.
Subagents Can Now Spawn Subagents, Three Levels Deep
Claude Code loosened a key constraint on agent orchestration: subagents can now spawn nested subagents up to depth 3 by default, up from a single level. Teams that want the old behavior can cap it with the CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH environment variable. It opens the door to genuinely hierarchical task decomposition — a planner delegating to workers that delegate to specialists.
July’s wider Claude Code haul rounds it out: in-app browsing, /doctor and /fork commands, public artifact sharing, artifacts that can reach each viewer’s own MCP connectors, and editor roles — plus a Windows fix that stopped paths like C:\Users\unicorn from being mangled into CJK characters.
MCP Crosses 400M Monthly Downloads as Enterprise-Managed Auth Arrives
The Model Context Protocol keeps compounding. It just crossed 400M monthly SDK downloads — a roughly 4× jump this year — and by Anthropic’s count, 78% of enterprise AI teams now run MCP-backed agents in production, with 28% of the Fortune 500 running their own MCP servers.
New enterprise-managed auth is the adoption unlock: admins can provision MCP connectors org-wide through their identity provider, authorize a connector once, and let users inherit access through existing IdP groups — connected on first login, zero-touch for the end user. The protocol Anthropic open-sourced is now the industry’s default plumbing for wiring agents to real systems.
AI for Science Grants Offer $50K in Claude Credits for Rare-Disease Research
Anthropic’s AI for Science program is offering grants of up to $50,000 in Claude credits over six months to researchers working on rare genetic diseases. The credits are meant to put frontier models directly in the hands of scientists tackling problems that rarely attract commercial funding.
Timing matters: applications for this cycle close today, August 2. It’s a small program in dollar terms, but it fits a pattern of Anthropic seeding Claude into scientific workflows — genomics, literature synthesis, and hypothesis generation — where a capable reasoning model can meaningfully compress the research loop.
Claude Code’s 50% Weekly Usage Boost Extended Through August 19
A bit of breathing room for heavy users: Anthropic extended the temporary 50% weekly usage boost for Claude Code subscribers, now running through August 19, 2026. Subscribers get meaningfully more headroom on their weekly limits before the boost expires.
It’s a modest quality-of-life move, but a telling one. With agentic coding sessions burning through tokens far faster than chat ever did, usage caps have become the real friction point — and Anthropic keeps nudging the ceiling up rather than letting limits become the reason developers drift to a competitor.
A Plumbing Week — and Plumbing Is Where Platforms Are Won
No new model today. Instead: a pricing deadline, a rate-limit bump, tiers folded from many into three, API keys that expire, subagents that nest, managed auth for MCP, and another top-tier integrator signed. On their own, each is a footnote. Together they describe a company operationalizing — smoothing the sharp edges that make a platform annoying to run at scale.
That’s worth paying attention to precisely because it’s boring. The Sonnet 5 pricing change is the tell: promotional rates were about winning trials; the September step-up, paired with a denser tokenizer, is about monetizing workloads that are now sticky. Meanwhile, matching Sonnet and Haiku limits to Opus, simplifying tiers, and shipping key expiration all lower the friction of expanding an existing deployment. The models grab the headlines, but weeks like this — unit economics, governance knobs, and distribution deals — are how an AI lab turns a demo into infrastructure that’s hard to rip out.