July 17
OpenAI opens the GPT-5.6 family to the public — the same day Moonshot launches Kimi K3, the world's largest open-weight model.
Curated by Thiago Lourenço Martins
OpenAI opens the GPT-5.6 family — Sol, Terra, and Luna — to the public Critical
After the White House asked it to slow the rollout over safety concerns on June 25, OpenAI began the broader release of the GPT-5.6 family — three models (Sol, Terra, and Luna) for different needs and budgets.
Access to a top-tier model (Sol) for complex tasks and more affordable options (Terra and Luna) for everyday work — Terra delivers performance close to GPT-5.5 at half the price.
Before switching AI providers, test all three GPT-5.6 tiers on real tasks from your operation — the savings come from using the right model for each task, not the most expensive one available.
[UPDATE] Moonshot AI launches Kimi K3, the world's largest open-weight model Critical
China's Moonshot AI launched Kimi K3, a 2.8-trillion-parameter open-weight model — the largest ever released publicly — which tied technically with Claude Fable 5 in code optimization and beat GPT-5.6 Sol and GPT-5.5 on independent benchmarks (Vals AI, Artificial Analysis).
The frontier of open, commercially free AI keeps getting closer to what closed labs deliver.
Technical teams can test Kimi K3 via API for complex coding at a fraction of the cost before renewing annual contracts with closed providers.
Meta negotiates renting US$ 10 billion in computing power for Anthropic — its direct AI rival High
According to the New York Times (confirmed by Reuters and CNBC), Meta is negotiating to rent up to US$ 10 billion in data center capacity to Anthropic over two years — the same Meta that competes directly with Anthropic on AI models.
The scarcity of compute power is rewriting the rules of competition: even direct rivals are becoming suppliers to each other.
If your company has idle cloud capacity, Meta's move is a reminder that it can become revenue — it's worth mapping what's spare and who sees value in it.
[UPDATE] Google delays Gemini 3.5 Pro again — Alphabet loses US$ 200 billion in market value High
Google delayed Gemini 3.5 Pro again (which had July 17 as its target) after the model failed to meet internal code-generation benchmarks; Alphabet shares fell about 4%, wiping out nearly US$ 200 billion in market value.
Even giants with virtually unlimited cash and talent hit the same bottleneck as smaller startups: making a frontier model shine at code.
Teams that tie roadmaps to third-party releases should always have a plan B — multi-month delays on the "right" model are the rule in this AI cycle.
Nadella publicly criticizes Anthropic's Claude Fable content restrictions Medium
Microsoft CEO Satya Nadella called Claude Fable 5 "editorially controlled" and argued that companies need to own their own models instead of "renting" capacity from a handful of companies — criticism coming from within a US$ 5 billion partnership between Microsoft and Anthropic.
A sign that even the biggest AI customers want less dependence on a handful of closed providers, pushing investment toward proprietary and open-source models.
When choosing an AI provider for your product, evaluate not just model quality but how much its content policies could block legitimate use cases for your business.
European Union forces Google to open Android and search data to AI rivals High
The European Commission ruled that Google must give rival AI assistants (ChatGPT, Claude, Perplexity) the same access to Android that only Gemini has today, as well as share search data — with deadlines of January 2027 (data) and July 2027 (Android).
It could open the first mass-distribution channel for AI assistants competing with Gemini across billions of Android devices in Europe.
Brands and developers of AI assistants should closely track the technical rules the Commission will detail — it could become an unprecedented user-acquisition channel in Europe.
DeepSeek prepares funding round at US$ 74 billion valuation ahead of Shanghai IPO High
DeepSeek is preparing a new round at a valuation of about US$ 74 billion (500 billion yuan), seeking to raise up to 50 billion yuan ahead of a possible IPO on the Shanghai stock exchange.
It reinforces that the market is betting heavily on Chinese open-AI labs, keeping low-price pressure on Western competitors.
Companies budgeting AI for 2027 should factor in that the US-China price war still has legs — negotiating multi-year contracts now could lock in better terms.
Focused on code and agentic tasks — xAI says it's the most significant capability jump since the original Grok launch.
→ x.aiA judge denied a block request from employees who claim they were laid off because of an AI-driven selection process.
→ reuters.comThe AI coding startup hit unicorn status just over a year after launch.
→ techcrunch.comThe company filed a lawsuit against a user who used Grok to generate child sexual abuse material (deepfakes).
→ reuters.comGoogle's conversational search now lets you link and interact with select apps directly within results.
→ techcrunch.comGet Radar IA every day on WhatsApp.