Weekly Recap · Week 27/2026

AI Recap: Claude Sonnet 5 becomes the default, Fable 5 returns — and the US government wants a cut of OpenAI

A week in which good AI becomes cheap and everyday: Anthropic makes the agentic Claude Sonnet 5 the default model for everyone, brings Fable 5 back after the export ban, and Google clears Gemini 3.5 Pro for July. Meanwhile, following the money shows how much is at stake: OpenAI offers the US government a 5% stake, and the first half of 2026 breaks every funding record at $510 billion. The week of June 29 – July 5, 2026 — in brief, with sources.

The week in one sentence

Frontier-grade AI slides into the mid-tier: Claude Sonnet 5 becomes the agentic default model for everyone — at half the Opus price —, Fable 5 is back after the US export ban, and Google clears Gemini 3.5 Pro. How much is at stake shows in the finance headlines: OpenAI offers the US government a 5% stake, and two labs capture 43% of all global startup investment this half-year. The lesson stands: whoever picks models flexibly wins.

Claude Sonnet 5 becomes the new default — agentic and far cheaper

Anthropic released Claude Sonnet 5 and made it the default model for Free and Pro users straight away (also available to Max, Team and Enterprise). The model works noticeably more agentically than Sonnet 4.6 — it checks its own output, finishes complex tasks and, at higher "effort", reaches Opus 4.8 levels on some tasks. The kicker is the price: $2 / $10 per million tokens (input/output) through August 31, then $3 / $15 — well below Opus 4.8's $5 / $25. For SMBs that means: your default model now handles near-autonomous work at a mid-tier rate. What that delivers in practice is in our Sonnet 5 for SMBs guide; how to turn it into real automation, our piece on AI agents for SMBs.

Source: Anthropic Source: VentureBeat

Fable 5 returns globally — with a new jailbreak framework

The Fable 5 saga gets its provisional happy ending: after the US export restriction was lifted, Anthropic redeployed Fable 5 globally on July 1 — with hardened cyber safeguards and a framework, developed together with Amazon, Microsoft and Google, that scores the severity of jailbreaks. After weeks of back-and-forth, Anthropic's model class is fully available again. The lesson from recent editions still holds, though: betting on a single, geopolitically vulnerable provider is a real operational risk — which alternatives are worth it, our AI tools comparison shows.

Source: Anthropic Source: 9to5Google

Google clears Gemini 3.5 Pro for July — and gets serious about agents

Google cleared the way for Gemini 3.5 Pro: after a delay, the launch is now set for July, while GPT-5.6 stays in the holding pattern and Fable 5 returns. In parallel, Google is expanding its agent lineup — with Gemini Spark, an agent that keeps working on tasks around the clock in the background, and Managed Agents in the Gemini API that autonomously plan, code and browse the web inside sandboxed environments. For businesses, this makes "agentic AI" tangible: not chatting, but getting tasks done — a field we unpack in our piece on AI agents for SMBs.

Source: TechTimes Source: Google

OpenAI offers the US government 5% — AI becomes a state affair

According to CNBC and the Financial Times, OpenAI offered the US government a 5% stake in the company — modeled on the Alaska Permanent Fund, so the public shares in AI's upside. At the latest $852B valuation that's around $42.6 billion; Anthropic, Google and Meta are expected to cede similar stakes via a government vehicle. It's all "conceptual" for now and would likely need an act of Congress. For SMBs the message is less the number than the signal: the big providers are becoming politically contested infrastructure — and that's a regulatory matter here too, see our EU AI Act guide.

Source: CNBC Source: CNN

Record half-year: $510B in venture capital — 43% of it to two labs

The Crunchbase half-year report shows the scale: a record $510 billion flowed into startups worldwide in H1 2026 — more than in all of 2025. What stands out is the concentration: OpenAI and Anthropic alone raised $217 billion — 43% of all global startup capital. AI companies drew more than 70% of Q2 capital. For small companies that's both a warning and a tailwind: market power is clustering with a few providers — which makes a multi-provider strategy all the more important, so you're not tied to a single roadmap (more in the AI tools comparison).

43% of all global startup investment in the first half of 2026 went to OpenAI and Anthropic alone — $217B of $510B.Crunchbase, July 2026
Source: Crunchbase Source: Value Add VC

Sources & further reading

As of July 6, 2026. Figures on models, prices, stakes and directives are per the companies and the cited media, without warranty. Benchmarks and provider claims are, where flagged, reported by the vendors and should be verified independently.

Frequently asked questions about the AI week

What's new in Claude Sonnet 5?
Claude Sonnet 5 has been the default model for Free and Pro users since June 30, 2026. It works far more agentically than Sonnet 4.6 — it checks its own output and, on demanding tasks, reaches Opus 4.8 levels at higher effort. Introductory API pricing is $2 per million input and $10 per million output tokens through August 31, 2026, then $3 and $15 respectively.
Is Claude Fable 5 available again?
Yes. After the US export restriction was lifted, Anthropic redeployed Fable 5 globally on July 1, 2026 — with additional cybersecurity safeguards and a jailbreak-severity scoring framework developed together with Amazon, Microsoft and Google.
Do I have to label AI content from August 2026?
From August 2, 2026, the EU AI Act's transparency obligations apply: deepfakes, AI-generated text on matters of public interest and chatbot interactions must be labeled as such. A voluntary EU code of practice helps with practical implementation. What that means specifically for small and medium-sized businesses is summarized in our EU AI Act guide.
Why does the US government want a stake in OpenAI?
In early July 2026, OpenAI proposed that the US government take a 5% stake in the company — modeled on the Alaska Permanent Fund, so the public shares in AI's upside. The stake would be worth around $42.6 billion; other US labs are expected to cede similar stakes. The talks are conceptual so far and could require an act of Congress.

Don't want to miss any AI development?

We keep an eye on the AI world for you — and translate what actually matters for your marketing, your tools and your automation.

Talk to us