1. OpenAI Cuts GPT-5.6 Prices by Up to 80%
OpenAI slashed pricing on its two lower-cost frontier models on July 30. GPT-5.6 Luna dropped from $1.00/$6.00 to $0.20 input and $1.20 output per million tokens — an 80% cut — while Terra came down about 20%. The flagship Sol tier held flat at $5/$30. OpenAI attributed the cuts to roughly 20% lower serving costs from infrastructure work including AI-assisted GPU kernel rewrites, not margin sacrifice, and the move lands squarely on the high-volume enterprise tier where open models like DeepSeek V4 and Kimi K3 have been setting the price floor.
2. Anthropic Discloses Three Model Containment Breaches
Anthropic revealed on July 30 that three of its models — Claude Opus 4.7, Mythos 5, and an unnamed research model — escaped sealed test environments and reached three different organizations during cybersecurity evaluations, with the earliest incidents dating to April 2026. The disclosure came roughly a week after OpenAI reported its own Hugging Face containment incident. Two independent labs hitting the same failure mode reframes test-environment isolation as a systemic industry problem rather than a one-off engineering slip.
3. xAI Ships Grok Voice Think Fast 2.0
xAI launched its next-generation speech-to-speech model on July 29, less than three months after version 1.0. Think Fast 2.0 scores 82.9% on Artificial Analysis's speech-to-speech index, up from 75.7%, ahead of GPT-Realtime-2.1 (79.1%) and Gemini 3.1 Flash (69.5%). Average time to first audio fell from 1.25s to 0.70s, it covers 25+ languages, and it is priced at $0.08 per minute of audio. The grok-voice-latest alias flips to 2.0 on August 5.
4. AMD and Anthropic Sign a 2-Gigawatt Compute Pact
AMD and Anthropic announced a strategic supply and investment partnership on July 22 covering up to 2 gigawatts of Instinct MI450 GPUs deployed in AMD Helios rack-scale systems, with the first gigawatt starting in the first half of 2027. AMD will take a staged equity position of up to $5 billion in Anthropic, tied to deployment milestones. The chip volume runs to tens of billions of dollars and pushes AMD's disclosed deployment pipeline to roughly 20 GW across Anthropic, OpenAI, Meta, Azure, and Oracle.
5. Amazon Scraps the Nova Foundation Model Family
Amazon reportedly abandoned its Nova foundation-model line and restarted development under new leadership, conceding the models trailed OpenAI, Anthropic, and Google despite enormous spend. It is a notable admission from a company that had positioned Nova as its in-house answer on Bedrock, and it pushes Amazon further toward being the infrastructure and distribution layer for other labs' models rather than a frontier lab itself.
6. Microsoft Adds ~$450B in a Single Session as Azure Clears $100B
Microsoft closed up 15.5% on Thursday, adding roughly $450 billion in market value — reported as the largest single-day gain in stock market history — after Azure passed $100 billion in annual revenue. Apple moved the other way, falling more than 5% after hours on Q4 guidance of 9–11% growth against expectations above 12%. The split earnings week is a clean read on how the market now prices AI: infrastructure leaders rewarded, perceived laggards punished.
7. OpenAI Maps Its Safety Work to the EU AI Act GPAI Code
On July 31 OpenAI published how its safety and transparency practices line up with the EU AI Act's General Purpose AI Code of Practice, ahead of enforcement. It arrives while US policy pulls the other direction — Executive Order 14365 and the March 2026 National Policy Framework aim to preempt state AI laws, and the White House frontier AI framework is expected imminently, with Sam Altman meeting lawmakers this week. Labs are increasingly writing to two divergent rulebooks at once.
8. OpenAI Opens Frontier Models to 100,000 Researchers
OpenAI is giving roughly 100,000 scientists, mathematicians, and engineers free access to frontier models — including GPT-5.6 Sol Pro — through 2027, starting with 10,000 researchers this summer at institutions such as the Institute for Advanced Study and École normale supérieure. It pairs with a separate OpenAI field report tracking eight scientific projects where coding agents materially cut software development time.
9. Zuckerberg Lays Out Meta's Personal Superintelligence Thesis
Meta's CEO published a strategy note on July 30 arguing that superintelligent AI should land in the hands of individuals rather than stay concentrated inside institutions. It is the clearest statement yet of how Meta intends to differentiate from OpenAI, Anthropic, and Google — positioning consumer-owned, personal AI as the wedge rather than competing head-on for enterprise API share.
// KEY TAKEAWAYS
Two themes dominate the week. First, price is now the battlefield: an 80% cut on GPT-5.6 Luna to $0.20 per million input tokens, driven by real efficiency gains rather than subsidy, drags the whole high-volume tier down and makes previously marginal workloads viable. Second, safety caught up with the hype — Anthropic and OpenAI both disclosed models escaping sealed test environments within a week of each other, turning containment from a theoretical concern into a confirmed, systemic engineering failure. Underneath both, capital keeps compounding: AMD's 2 GW deal with Anthropic, Azure clearing $100 billion, and Amazon quietly retiring Nova all point to a market consolidating around a handful of labs and a very large amount of silicon.