The top things worth knowing about in AI today.
SpaceXAI opened public access to Grok 4.5 on grok.com and X, a model built for coding and agentic work. It lands fourth on the Artificial Analysis intelligence index and costs $2 per million input tokens and $6 per million output, roughly 60% below Claude Opus 4.8 or GPT-5.5. Musk rates it close to Opus 4.7 but faster, giving developers a cheaper option for high-volume code and research tasks.
Read more →Google pushed the release of Gemini 3.5 Pro to July 17, scrapping its 2.5 Pro architecture for a full rebuild aimed at maths, reasoning and image quality. The delay followed a week in which four senior DeepMind researchers left, including Gemini co-lead Noam Shazeer to OpenAI and Nobel laureate John Jumper to Anthropic. Alphabet lost around $225 billion in market value on the news. For teams planning around Gemini, the frontier timeline has slipped again.
Read more →Meta released Muse Spark 1.1 and opened the Meta Model API, the first time it has charged businesses to use one of its models rather than releasing the weights openly. Zuckerberg said it tests better than Google's Gemini on agents, coding and multimodal tasks, and priced it at around 25% of what OpenAI and Anthropic charge. Developers start free and pay above a token threshold. It marks Meta trying to earn revenue from AI spending it has struggled to justify internally.
Read more →Anthropic rolled out Reflect, which shows how much you have used Claude and on what, with summaries of key topics and patterns over the past 1, 3, 6 or 12 months. It is a small addition but a useful one for anyone trying to work out where AI actually fits into their week. Teams can use it to see which tasks are moving to Claude and which are not.
Read more →Microsoft added visual and audio watermarks to AI-generated content across Microsoft 365 Copilot, with an admin policy controlling whether an organisation applies them. The change is meant to make clear when video or audio has been generated or altered by AI. For businesses worried about provenance and disclosure, it is a practical control rather than a policy promise.
Read more →The Australian government began a five-year sourcing agreement with Microsoft on 1 July that gives public service agencies access to Copilot, Microsoft 365, Azure and Dynamics 365 at fixed pricing. Separately, Wesfarmers signed a multi-year deal with Google Cloud to run agentic AI across Kmart, Officeworks, Priceline and OnePass. With 42% of Australian small and medium businesses already using AI tools, the technology is moving from trials into standard government and retail operations.
Read more →Global venture funding hit $510 billion in the first half of 2026, more than the $440 billion invested in all of 2025, with over 70% of second-quarter capital going to AI companies. Anthropic raised $65 billion in the quarter to become the most valuable private company, and together with OpenAI took 43% of all startup funding. The money is concentrating in a handful of frontier labs, which shapes who can afford to train the next generation of models.
Read more →A study found companies spending around $30 per employee a month on AI grew headcount 10% over two years, while low adopters saw little change. Entry-level hiring rose 12% at the heaviest adopters, cutting against the assumption that AI is thinning junior ranks. Separate research shows no rise in aggregate unemployment for exposed workers since 2022, so the shift so far is in what jobs involve, not whether they exist.
Read more →A short read every week — the few things worth your time, and nothing that isn’t.