Meta shut the door on open-source AI for good, Chai Discovery raised $400M to let AI design drugs, and the EU's AI law is 11 days from enforcement.
Meta kills Llama. Muse Spark 1.1 is now a paid, closed model.
Meta has officially ended its open-weight Llama era. Muse Spark 1.1 (built by Meta Superintelligence Labs) launched July 9 as the company's first proprietary, closed AI model: no public weights, no open license, API access only at $1.25 per million input tokens and $4.25 per million output tokens. This is the same Meta whose CEO spent three years publishing manifestos about open AI and criticising OpenAI for going closed.
Why it matters: When the world's loudest open-source champion goes proprietary, it changes the math for developers, students, and small businesses who relied on free Llama weights. If you built anything on open Meta models, watch this closely.
Frontier AI just got 60% cheaper. Here's what the price war looks like now.
July 2026 produced a rare moment in tech: output token costs for frontier AI fell from $25–50 per million tokens to $4–6. Grok 4.5 (launched July 8 at $6/million output) and GPT-5.6 Luna ($6/million) undercut previous flagships by over 60%. Grok 4.5 also runs at roughly 80 tokens per second with a 500K context window, and benchmarks show it doing the same work as older models while using 2–4x fewer tokens.
Why it matters: Cheaper inference means AI tools your college or small employer couldn't afford last year may now be within reach. The price floor for serious AI just dropped significantly.
Chai Discovery raises $400M to let AI design the drugs of the future
San Francisco-based Chai Discovery closed a $400M Series C on July 14, tripling its valuation to $3.8B. The company, founded in 2024, builds AI models that predict and reprogram how molecules interact, accelerating early drug discovery before clinical trials begin. Its platform is already deployed at Eli Lilly and Pfizer. The round was led by Index Ventures alongside Kleiner Perkins and Sequoia.
Why it matters: $630M raised in under two years for a company that didn't exist in 2023. AI for drug discovery is no longer speculative, it's being written into pharma pipelines at the world's largest companies.
Fake AI job candidates are now a documented crisis. The US just gave victims a way to fight back.
Pindrop Security analysed 300 job applicant profiles and found over one-third were entirely fabricated using AI. Gartner projects 1 in 4 candidate profiles globally will be AI-generated fakes by 2028. The US DEFIANCE Act, passed unanimously by the Senate in January, creates a federal right for victims of AI-generated synthetic fraud to sue creators and distributors, moving enforcement from a grey area into law.
Why it matters: If you're a hiring manager, this is already your problem. If you're a job seeker, companies are now screening for authenticity in new ways. Both sides of every job interview are changing.
The EU AI Act is 11 days from full enforcement. Most businesses aren't ready.
Europe's landmark AI law enters full enforcement on August 2, 2026. Requirements include mandatory transparency labels on AI-generated content, prohibited use cases for high-risk AI systems, and compliance obligations for companies that deploy AI in hiring, credit, education, and healthcare. Enforcement applies to any company serving EU customers, not just European ones.
Why it matters: If your company uses AI in any decision-making process affecting EU users and hasn't checked compliance, the clock is nearly out. Non-compliance fines can reach €35M or 7% of global annual turnover.
Google DeepMind and NASSCOM launch a free 56-hour AI course for India
Google DeepMind has launched its AI Research Foundations curriculum (a free 56-hour programme teaching learners how to build and fine-tune large language models) in partnership with NASSCOM, the Indian Institute of Science (IISc) in Bengaluru, and other institutions across India. Separately, Emergent, an Indian AI startup, became the country's latest unicorn, and the government unveiled a Rs 1.27 lakh crore semiconductor plan to reduce dependence on imported chips.
Why it matters: A free, IISc-backed AI curriculum is a rare resource for students and professionals who want real technical depth without a premium price tag. This is worth bookmarking and sharing.
What happens to AI access when the price collapses
When output tokens cost $50 per million, frontier AI was for well-funded companies and research labs. At $4–6, the same capability lands in the budget of a school project, a solo founder, or a small NGO. The July 2026 price war is still playing out, but the more interesting question isn't which model wins, it's what gets built now that the cost barrier has been lowered by 90%. The history of the internet suggests cheap access unlocks uses nobody in the labs predicted.
Databricks raises $3B at $188B valuation + recent grads face a harder job market
Databricks closed a $3B round at a $188B valuation, cementing its position as the infrastructure layer underneath much of enterprise AI. On jobs: a Federal Reserve Bank of New York study found recent college graduates (ages 22–27) face 5.6% unemployment versus the 4.2% overall rate: the gap is widening even as AI job postings surge 600% over three years. The tools are growing; the entry path is narrowing.