AI Trends — 17 August 2026

The price war intensifies

The past 48 hours brought a fresh round of AI pricing moves. OpenAI cut GPT-5.6 "Luna" input pricing by 80%, to $0.20 per million input tokens. Google countered with Gemini 3.7 Flash, priced at $0.75 per million input tokens and $3.75 per million output tokens — roughly half the cost of the prior Gemini 3.6 Flash generation, with the promotional pricing set to run through the end of 2026. DeepSeek moved in the opposite direction, raising prices on its V4-Pro and V4-Flash models effective August 16, with off-peak output pricing climbing from $0.87 to $1.98 per million tokens (and up to $3.96 during peak hours).

Open-weight releases keep coming

Meta released Muse Glimmer, a 30B-parameter dense multimodal model under an Apache 2.0 license, tuned for local agentic tool use, coding, and LLM-as-judge tasks, with a 131K context window and support for more than 100 languages. Alibaba's Qwen team followed with Qwen 3.8 27B, also Apache 2.0 licensed, featuring integrated vision and a native 262K context window that extends to 1M, built on a Gated DeltaNet plus Gated Attention architecture.

Google's AI leadership shake-up

In one of the bigger stories of the week, Jeff Dean and three other senior Google AI researchers announced they are leaving after 27 years at the company to launch their own startup. Google is reorganizing in response: Demis Hassabis is stepping back to a Chairman role, Koray Kavukcuoglu is taking over operations, and Sebastian Borgeaud's coding team is relocating from London.

Meta's bigger bet

Mark Zuckerberg published a roughly 6,500-word manifesto laying out Meta's AI ambitions, arguing for wider release of open-weight models and proposing ways the federal government could partner with AI companies on safety testing. Alongside it, Meta announced a $1 billion fund to invest in communities surrounding its data centers, and unveiled Muse Code, a new AI coding assistant built on its Muse Spark 1.2 model.

By the numbers

OpenAI confirmed it has crossed 1 billion users and is reportedly preparing for a public offering that could rank among the largest tech IPOs in history. Anthropic reported Q2 2026 revenue of $10.9 billion, up 130% year over year, and its first operating profit of $559 million — arriving two years ahead of the company's own internal schedule.

The takeaway

Three threads defined AI news this week: falling inference costs, a widening field of capable open-weight models, and a scramble for talent and positioning among the largest labs as commercial stakes keep rising.

Sources: