Stephanie Palazzolo

OpenAI's Codex tool is gaining traction among developers switching from Anthropic's Claude Code, driven by recent improvements for complex, longer-running tasks. The productivity gains have created an unintended consequence: OpenAI engineers are now submitting more than 10 code changes per day, up from two or three, overwhelming the company's internal testing infrastructure and causing outages in systems that manage its codebase.

OpenAI's foundation announced $250 million in grants to fund research on artificial intelligence's economic and labor market impacts. The charitable arm, separated from OpenAI's for-profit business last fall, will distribute the funding to advance understanding of how AI affects jobs and economic outcomes. The initiative signals OpenAI's effort to address growing concerns about AI's societal consequences through independent research.

OpenAI has been without a communications chief since December when Hannah Wong departed, and the company is struggling to fill the role as multiple high-profile candidates have declined or hesitated. Potential hires, including Uber's Jill Hazelbaker and others, have cited the difficulty of managing CEO Sam Altman and other executives who operate with significant autonomy in public-facing roles. The vacancy reflects a broader PR vulnerability for OpenAI as it heads toward a potential IPO amid growing political backlash against the AI industry and a competitive battle with Anthropic and Google.

As frontier AI model costs rise, some developers are exploring open-source alternatives like DeepSeek V4 and Moonshot AI's Kimi K2.6 to reduce expenses, with companies like Uber and Airbnb already shifting workloads to cheaper models for simpler tasks. However, early feedback suggests open-source models still lag on reasoning depth, performing well on benchmarks and surface-level questions but struggling with follow-up questions or deeper reasoning chains. While open-source adoption is growing overall based on inference provider data, it remains unclear whether these models can fully replace frontier offerings for complex use cases.

SiMa.ai, a San Jose-based startup developing low-power inference chips for edge devices like drones and cameras, is raising over $100 million at a $1.4 billion valuation, representing a 45% premium from its $960 million valuation last August. The funding reflects investor conviction that specialized AI chips for edge inference represent a distinct market opportunity separate from the data center GPU dominance. SiMa.ai is among several startups betting that not all future AI applications require massive centralized computing power, which could reshape infrastructure investment expectations.

Anthropic and OpenAI are consolidating market dominance in the AI startup sector, with the two companies now capturing 89% of revenues across a group of 34 leading AI startups. The broader cohort generated nearly $80 billion in annualized revenue, or $6.6 billion monthly, from AI applications and model access, representing 112% growth over six months. This widening gap underscores how quickly the market is concentrating around the two largest players while other startups struggle to compete.

OpenAI announced a $10 billion private-equity joint venture called the OpenAI Deployment Company, designed to help enterprises deploy and integrate AI into their operations. The company also acquired Tomoro, an AI consulting and engineering firm, as part of this expansion. The move signals OpenAI's shift toward enterprise deployment infrastructure and professional services alongside its core AI model business.

OpenAI has engaged Gimlet Labs to optimize its AI models for Cerebras chips as major AI developers diversify away from Nvidia's constrained GPU supply. The startup handles the technical grunt work of tailoring model code to run efficiently on alternative chip architectures. Gimlet's work with OpenAI includes optimizing models that power Codex-Spark, a faster version of OpenAI's coding tool. This arrangement reflects a broader industry shift where AI labs must now support multiple chip types to secure adequate computing capacity.

Rising costs for advanced AI models are now straining budgets even at large, well-resourced organizations. Uber burned through its 2026 AI budget in months, and venture capital firms are discovering that casual usage of premium models like Claude can cost $1,000 per day per user. The core issue is that many users default to the most advanced and expensive models for routine tasks, creating runaway costs that can exceed $100,000 monthly for small teams if left unchecked.

Despite SpaceX's $60 billion conditional takeover offer for Cursor last month, the coding startup is maintaining independence in its AI model strategy. Cursor does not plan to codevelop coding models with SpaceX's xAI unit and instead is focused on improving its own Composer model, which is partially powered by Chinese model Kimi. The startup also has no plans to direct customers toward xAI's Grok model for coding tasks, instead relying on Composer, Anthropic's Claude, and OpenAI's Codex as its primary model options.

A UK government group conducting AI cybersecurity testing has found that OpenAI's GPT-5.5 model performs comparably to Anthropic's unreleased Claude Mythos model on certain security tasks. In a difficult corporate network attack simulation, GPT-5.5 succeeded in 2 out of 10 attempts, matching Mythos performance levels. The finding suggests both leading AI labs have achieved similar capabilities in this specialized security domain, though the incomplete article limits full context on scope and implications.

Microsoft and other major cloud providers are restricting GPU availability to smaller AI startups by allocating Nvidia chips to their own internal teams and larger customers, creating a supply bottleneck that forces well-funded startups to pay higher prices for remaining capacity. The shortage is affecting companies backed by top-tier investors including Sequoia Capital, Founders Fund, General Catalyst, and Andreessen Horowitz, prompting at least one major VC firm to survey founders about compute access constraints.

OpenAI is testing a new image generation model, internally referred to as 'gpt-image-2,' that produces photorealistic images nearly indistinguishable from real photographs. The model is being tested with select ChatGPT users and on leaderboards, with examples circulating on X and Reddit. Beyond its technical capabilities, the release appears tied to OpenAI's broader strategy to reach 1 billion weekly active users on ChatGPT, a milestone the company missed in 2025 and has been pursuing since stalling at around 920 million WAU.

Anthropic is preparing Claude Opus 4.7, its next flagship language model, alongside a new AI-powered design tool for websites and presentations. Both products could launch as soon as this week, according to a source with knowledge of the roadmap. The releases would represent Anthropic's continued push to expand its model capabilities and extend AI tooling into creative and design workflows.