Mara Lindqvist covers the model providers, runtimes, and sandboxes that production agents are built on. Her beat takes in foundation-model releases that ship agent-relevant capabilities — tool use, computer use, long context — provider SDK changes, and the inference infrastructure underneath. She has reported on AI infrastructure for several years and prefers hard news to commentary.
The largest expansion of OpenAI's six-month-old ad business goes live August 24 across the EEA, initially through agency partners, with a self-serve Ads Manager for small advertisers still to come.
The August 24 launch is OpenAI's largest ad-market expansion to date and adds conversion optimization, geo-targeting, an OpenAI Pixel, and a Conversions API — but self-serve access for small advertisers is still pending.
The payments company confirmed the deal on August 19, calling routing across 400+ models a new layer of its 'economic infrastructure for AI.' The NYT put the price at about $7.5 billion, with $1.5 billion going to OpenRouter's founders.
The all-stock deal, effective August 14 per an 8-K filing, issued 389.3 million SpaceX Class A shares to Cursor holders and consolidates the coding-agent push behind Grok Bot, a persistent multi-agent workforce product now in beta.
The payments company has agreed to acquire the routing layer that funnels roughly 8 million developers to more than 400 AI models, months after OpenRouter last raised at a $1.3 billion valuation.
The Hangzhou lab is switching to peak/off-peak billing on August 16, with V4-Flash output rising from $0.28 to $1.32 per million tokens and cache-hit reads jumping roughly sixfold — the change that hits agent pipelines hardest.
The Palo Alto startup from former xAI co-founder Igor Babuschkin drew in General Catalyst, AMP PBC, Nvidia, AMD Ventures, Y Combinator, and Temasek on a thesis that enterprises — and eventually individuals — should own and retrain their own open-weight models.
Cloudflare Wallets and cloudflare.pay, launched Tuesday during Agents Week, pair a human-readable agent identity with Virtual Wallets that can autonomously buy APIs, MCP tools, and content — completing the buyer side of the two-sided market the Monetization Gateway opened in July.
Gemini 3.6 Flash cuts output tokens by 17% at $1.50/$7.50 per million, a 350-tokens-per-second 3.5 Flash-Lite targets high-volume subagents, and a government-only 3.5 Flash Cyber runs inside CodeMender — while the long-delayed 3.5 Pro remains in partner testing and Gemini 4 pre-training begins.
The Beijing lab's 2.8-trillion-parameter MoE — with a 1-million-token context window and full weights due July 27 — beat Claude Opus 4.8 and GPT-5.5 on coding and general-agent tasks, and sent Chinese AI rivals and US chip stocks sliding on Friday.
Powered by the newly released GPT-5.6, the cloud-based agent handles multi-step workflows across Gmail, Slack, Calendar, and GitHub — and takes direct aim at Anthropic's Claude Cowork ahead of dueling IPOs.
The three-tier frontier family went to general availability July 9 following a Commerce Department review that gated its release, alongside a new long-running enterprise agent called ChatGPT Work.
The new agent runs multi-hour workflows across connected apps and files; the accompanying Sol, Terra, and Luna models are live in ChatGPT, Codex, and the API.
On its second 'Content Independence Day,' the company split crawlers into Search, Agent, and Training — and opened a Monetization Gateway that charges AI bots in stablecoins over the x402 protocol.
The new mid-tier model, released June 30 and now the default for Free and Pro users, matches Opus 4.8 on a knowledge-work benchmark and closes the gap on agentic coding — at introductory API pricing 60% below the flagship.
The mid-tier release closes much of the agentic-coding gap with the flagship Opus 4.8 at roughly 40 percent lower list pricing, and arrives with Anthropic's IPO prospectus confidentially filed and its most powerful models still under Trump-administration restrictions.
The flagship Sol model, a new 'ultra' multi-agent mode and a tiered Terra/Luna lineup launched Friday under a White House-coordinated limited preview that OpenAI says should not become the long-term default.
A working paper from OpenAI's Economic Research team, published June 25, documents a rapid shift from chat to delegated agentic work — though every figure is drawn from OpenAI's own platform telemetry.
At AWS Summit New York on June 17, Amazon unveiled an agentic code-vulnerability service, a company-wide knowledge graph, general availability of the Bedrock AgentCore harness, and a Kiro iPhone app.
The Mythos-class model goes general availability with classifier guardrails that punt cyber, bio, chem, and distillation prompts to Opus 4.8 — while a sibling build with the cyber safeguards stripped stays inside Project Glasswing.
At WWDC 2026, Apple unveiled a rebuilt 'Siri AI' powered by a custom 1.2-trillion-parameter Google Gemini model and introduced an Extensions framework letting users set rival assistants as the iPhone default.
At WWDC 2026, Apple unveiled a Gemini-powered Siri AI with multi-step tasks, on-screen awareness, and a standalone app — but no launch in the EU or China.
The Financial Times, citing more than a dozen current and former employees, says the redesign rolls out in the coming weeks, with Canva and Booking.com among launch partners and Codex's weekly active users already up sixfold to 5 million.
The new flagship model adds dynamic workflows that orchestrate hundreds of parallel subagents in Claude Code, arriving 41 days after Opus 4.7 — and alongside a Series H that pushes Anthropic past OpenAI on paper.
On the eve of what is expected to be Tim Cook's final keynote, Apple is set to unveil a rebuilt Siri running on Google Gemini, a standalone chatbot app, agentic cross-app tasks, and App Store hooks for third-party agents.
At a two-day developer conference in San Francisco, the company introduced an open agent governance spec, OS-level execution controls, a PostgreSQL-compatible agent database, and Project Solara — its bet on agent-first hardware.
The developer conference kicked off May 6 in San Francisco with a slate of agent-infrastructure releases — multiagent orchestration, dreaming, webhooks, and Outcomes — plus self-hosted sandboxes in public beta and MCP tunnels in research preview.