News
3 min read
July 24, 2026
A short field guide to a month of massive releases, government intervention and billion-user milestones. This is what each release means in plain language and what to do about it.
The short answer
June 2026 was a month of huge product stories but perhaps more importantly, it was a month with huge implications for AI policy.
- Anthropic shipped Claude Fable 5 on 9 June, their most powerful model yet, only for the US Department of Commerce to block foreign access within days. Controls were mostly lifted on 30 June, and access was restored from 1 July. Fable 5 was included on standard Claude plans through mid-July, and from 20 July it stays included in Max and Team Premium plans, while Pro and Team Standard users move to a usage credits model. On 30 June, Anthropic also released Claude Sonnet 5, bringing Opus-level performance to a model that costs about one-third as much.
- OpenAI announced its next model family, GPT-5.6 Sol, Terra, and Luna, and crossed one billion monthly active users in May, the fastest any app has ever hit that milestone.
- Google expanded Gemini Spark to macOS, retired its original CLI in favour of Agentic 2.0, and started rolling out Gemini 3 Deep Think to AI Ultra subscribers.
- Microsoft used Build 2026 to announce Scout (its first "Autopilot" agent), unveil seven new AI models, and ship Copilot Cowork to general availability worldwide.
The theme of the month is that AI companies are building fast enough that governments are struggling to keep up. Read on for the details.
Results and takeaways:
- Anthropic: Claude Fable 5 (most powerful model, $10/$50 per million tokens), was blocked then restored for Australian users, and from 20 July stays included in Max and Team Premium plans while Pro and Team Standard move to usage credits. Claude Sonnet 5 (30 June): Opus-level capability at $2/$10 per million tokens, rising to $3/$15 from September.
- OpenAI: GPT-5.6 Sol (frontier reasoning), Terra (competitive at half the cost), Luna (fastest and cheapest). Limited preview with vetted US partners only. One billion monthly active users.
- Google: Gemini Spark on macOS (US-only, no Australian rollout date). Agentic 2.0 CLI replaces retired Gemini CLI. Gemini 3 Deep Think rolling out to AI Ultra.
- Microsoft: Copilot Cowork GA worldwide. Scout "Autopilot" agent. Seven new Microsoft AI models. Build 2026.
What's new from Anthropic in June? — Huge news for all Claude users this month.
Claude Sonnet 5 arrived on 30 June as a late-month inclusion. It's Anthropic's most capable Sonnet to date, designed to bring Opus-level performance on coding and agentic tasks at about one-third of the cost. Launch pricing is $2 per million input tokens and $10 per million output tokens, with standard pricing of $3/$15 kicking in from September.
But the big news for Claude users…
Claude Fable 5 launched on 9 June as Anthropic's most powerful model, priced at $10 input / $50 output per million tokens and built for multi-day, complex agentic tasks. Within days, the US Department of Commerce applied export controls blocking access for all foreign nationals, including Australians. The stated reason was national security. On 30 June, the controls were mostly lifted after Anthropic agreed to proactively detect and report security risks. Access was restored globally from 1 July, with Fable 5 included on Pro, Max, Team and select Enterprise plans at up to 50% of weekly usage limits.
That included window was originally set to close on 7 July, but Anthropic extended it twice, first to 12 July, then to 19 July. From 20 July, Fable 5 stays included in Max and Team Premium plans, while Pro and Team Standard users move to a usage credits model, softened by a one-time $100 credit.
So if you tried Fable and found it blocked, it's available again, and it's staying in subscriptions. Worth a test.
Source: anthropic.com
What's new from OpenAI in June? — Two things worth noting.
ChatGPT crossed one billion monthly active users in May, confirmed in June, making it the fastest app ever to hit that milestone. For context, it took TikTok six years to reach a billion. ChatGPT did it in roughly three.
OpenAI also previewed its GPT-5.6 model family, which landed in early July: Sol (frontier reasoning and agentic work), Terra (GPT-5.5-competitive at half the cost), and Luna (fastest and cheapest). Sol is in limited preview with selected US government-vetted partners and enterprises only. Broader availability is coming, though the US government's recent pattern of blocking frontier model access for non-US users is worth keeping in mind. This one may follow a similar path before it reaches Australia.
Source: openai.com
What's new from Google in June? — Three releases, each at a different point on the maturity curve.
Gemini Spark expanded to macOS with new integrations for Canva, Dropbox, Instacart and others. It's still US-only, with no Australian rollout timeline announced. Google's personal agent offering is also looking impressive, but most of us outside the US are still waiting to actually test it.
Agentic 2.0 CLI replaced the Gemini CLI on 18 June. The old CLI is retired.
Gemini 3 Deep Think began rolling out to AI Ultra subscribers as Google's top reasoning mode for complex multi-step problems. It's positioned as the reasoning layer on top of an already capable model. Early access only for now.
Source: blog.google
What did Microsoft ship in June? — Build 2026 saw more news from Microsoft.
Copilot Cowork reached general availability worldwide, the agentic workspace tool Microsoft announced last month is now shipping to all users. The tool is still fairly limited in what it can do autonomously and gets stuck on tasks that a more mature agentic tool handles cleanly.
Scout is Microsoft's first "Autopilot" agent, an AI that can operate independently across tasks without step-by-step instruction. It's an early preview, but it signals that Microsoft is heading away from Copilot as a chat assistant and toward Copilot as an autonomous layer, weaved into your existing processes.
Microsoft also previewed seven new AI models covering coding, reasoning, image, and voice.
Microsoft continues building its own model capability, which should mean more choice (and more complexity) for the organisations using their stack.
Source: microsoft.com
What's the pattern across all four?
June was all about what happens when the models get good enough that governments are forced to act.
The US Department of Commerce blocked Fable 5 within days of launch. GPT-5.6 Sol is gated behind government vetting. This is proving a real constraint for Australian organisations that want to run on the most capable tools as soon as they're announced.
Inside those constraints, the technical direction is consistent. We're seeing more agents that operate without step-by-step supervision (Scout, Fable 5, Gemini Spark). Models with better economics (Sonnet 5, Terra). Personal agents that work across your full digital life.
What should you do about it?
Take access seriously. If your AI strategy depends on frontier models, build in contingency for access interruptions. June showed that a model can be available one day and blocked the next.
Sonnet 5 is worth a solid look. For most teams, this is the most practically useful release of the month. Opus-level performance at a fraction of the cost, from a vendor that's been consistent on quality.
The agent gap is widening. Microsoft's Copilot Cowork is GA but a little limited. Gemini Spark is impressive but US-only. Claude and OpenAI are the practical options for most Australian organisations right now.
What we're watching in July:
- Will Fable 5 access stay stable for Australian users, or will the export controls return in a different form?
- When does Gemini Spark actually arrive in Australia?
- GPT-5.6 Sol is previewing with vetted partners. What do the early results actually look like?
- Microsoft's Copilot Cowork has GA status. Will the product catch up to the label?
If you'd like to talk about what any of this means for your team, that's the work we do. AITC runs hands-on training and advisory engagements for organisations that want their teams to use AI safely and effectively. Get in touch.
