News

v2.44.0 was released on 2026-06-24 with a bug fix in auth to prioritize the first auth header. This matters because header precedence can affect which credentials are used when multiple auth headers are present, reducing ambiguity and unexpected authentication behavior.

v0.112.0, released on 2026-06-24, adds client support for streaming `system.message` events and includes API updates for a new refusal category and sending a User Profile ID in request headers. The release also fixes the memory tool so it creates parent directories with the correct permissions, a small but important filesystem-safety change.

Google introduced computer-use capabilities in Gemini 3.5 Flash, enabling the model to interact with computer interfaces as part of its workflow. It matters because this moves Gemini from text and image generation toward agentic task execution, a step toward automating multi-step actions in software.

NVIDIA NeMo AutoModel is being highlighted for accelerating Transformer fine-tuning workflows. It matters because faster fine-tuning can shorten iteration cycles for adapting large models to new tasks and datasets.

OpenAI revealed its first AI server chip, Jalapeño, an “intelligence processor” built with Broadcom and designed as an ASIC for AI inference in current and future large language models. It matters because it marks OpenAI’s move into custom silicon for serving products like ChatGPT and Codex, following its chip partnership announcement just nine months earlier.

xCures raised a $46 million Series B led by Innovius Capital, with iGrow, Spring Mountain Capital and existing backers participating, bringing total funding to more than $76 million and valuing the 2018 startup at $127 million post-money. The company, which pivoted from direct-to-consumer cancer decision support after wrestling with faxed and FedExed records, is betting its “Clinical Clarity Engine” can turn messy patient data into decision-ready clinical intelligence faster than transport-focused rivals.

Mike Schroepfer, Meta’s former CTO, founded Gigascale Capital in 2023 and just raised a $250 million first institutional fund to back companies rebuilding the physical economy, with more than 25 portfolio investments and check sizes from $1 million to $10 million. He says AI is making software cheaper while shifting the moat to infrastructure like power, compute, manufacturing and supply chains, and argues the combination of rising demand and falling costs in areas like solar, batteries and electrolyzers makes this a rare moment for hard-tech startups.

OpenAI and Broadcom unveiled Jalapeño, a custom AI chip designed for LLM inference to improve performance, efficiency, and scale across AI systems. A chip optimized specifically for inference could lower serving costs and increase throughput for large models, which is increasingly important as deployment demand grows.

FFASR introduced a new leaderboard for benchmarking automatic speech recognition in real-world conditions. It matters because it gives researchers and practitioners a way to compare ASR systems on practical, noisy, and diverse audio rather than only on controlled test sets.

Menlo Ventures raised $3 billion in new capital, its largest fundraise in 50 years, split between Menlo Ventures XVII for seed and Series A deals and Menlo Inflection IV for Series B+ AI startups across enterprise, healthcare, and consumer. The firm pointed to its early 2023 investment in Anthropic—now valued at $965 billion and planning a 2026 IPO—as the basis for its AI strategy and a potential path to its biggest exit ever.

GPT-5 Pro helped immunologist Derya Unutmaz solve a three-year-old mystery about T cell behavior. The result matters because it could advance cancer and autoimmune research by providing new insights into immune system regulation.

Stockholm-based Fika Jobs raised $4 million to build a video-first hiring platform that combines AI interview agents with short-form video profiles, aiming to blend LinkedIn-style recruiting with TikTok-like presentation. The approach matters because it shifts early candidate screening to automated interviews while giving employers a more visual, faster way to assess applicants.

CUGA introduces a lightweight harness for building agentic applications, with two dozen working examples demonstrating the approach. It matters because the examples suggest a practical way to prototype and ship agent workflows without a heavy framework stack.

Transformers.js is experimenting with the proposed Cross-Origin Storage API to enable cross-origin access patterns for browser-based ML workloads. This matters because it could simplify loading and sharing model assets across origins, but the source excerpt provides no implementation details or performance data.

Nvidia says its Rubin generation reference design for a fully liquid-cooled AI data center runs hotter while eliminating “massive amounts of power usage” and “pretty much all water usage.” The claim matters because data center water and energy use has become a major public concern, but Nvidia still doesn’t address construction impacts, power-generation demands, or the cost versus air-cooled designs.

Alan Greenspan, who died at 100, is being revisited for coining “irrational exuberance” in a 1996 speech and for his 1999 “lottery ticket” analogy about internet stocks. The piece argues those warnings aged well for the dot-com bubble, while drawing a parallel to today’s AI boom, where money-losing companies are again attracting huge valuations even as winners like Google and Amazon ended up worth nearly $8 trillion combined.

AppsFlyer reportedly raised more than $1 billion in a Series E at a $2.7 billion post-money valuation, with Moloco, Google, Meta and Unity each taking minority stakes. The marketing analytics company, founded in 2011 and now at $1.3 billion in known funding, is positioning itself as an independent attribution layer for digital ads and says the round is a step toward a public listing.

SpaceX has inked a compute deal with Reflection AI, which will pay $150 million a month starting July 1, 2026 through 2029 for immediate access to Nvidia’s latest GB300 AI chips and supporting hardware at SpaceX’s Colossus 2 data center near Memphis, Tennessee. The deal highlights how scarce top-end AI compute has become, with a long-term commitment worth about $1.8 billion a year to secure cutting-edge GB300 capacity.

AI agents are undermining traditional SaaS by reducing the need for per-seat subscriptions, with the piece pointing to a January $300 billion single-session wipeout and arguing that software is shifting toward AI-native products that charge by usage or outcomes. It says horizontal SaaS categories like form builders, project management tools, SMB CRMs and social schedulers are vulnerable, while vertical specialists with distribution, domain expertise and proprietary data moats are better positioned to capture the $2 trillion white-collar services market.

Robotics startups have already raised $18.8 billion globally in 2026, surpassing the full-year 2025 total of $15 billion and the 2021 peak of $14.1 billion, with major rounds including Saronic’s $1.75 billion Series D, Neura Robotics’ up to $1.4 billion Series C, Skild AI’s $1.4 billion round, Shihang Intelligent’s $1 billion Series A, and Mind Robotics’ $900 million across two rounds. The surge shows venture investors now see embodied AI and real-world robotics as a big opportunity, while exits remain quieter in the U.S. even as Chinese listings like Unitree and Robotphoenix point to a more active public-market path.

OpenAI introduced Daybreak tools, including Codex Security and GPT-5.5-Cyber, to help organizations find, validate, and patch vulnerabilities at scale. The new tools are meant to bring AI-assisted security workflows to every organization, with a focus on faster vulnerability discovery and remediation.

Seedcamp closed its 7th fund at $220 million and a separate $100 million select fund, bringing total assets under management to $1 billion after nearly two decades of investing since its 2007 launch. The firm, which has backed around 550 companies including Revolut, Wise, UiPath, Synthesia and Fluidstack, plans to make 100 to 120 new investments while focusing more on AI, robotics, defense and health as it expands its New York presence.

Jason Liu describes using Codex to preserve context and manage complex projects so work can continue beyond a single prompt. The key point is that Codex is being used for long-running tasks where retaining state and continuity matters more than one-shot answers.

Local models were used to triage the OpenClaw repo at no cost, according to the source excerpt. This matters because it suggests offline or on-device models can handle repository triage without relying on paid hosted inference.

Samsung Electronics is deploying ChatGPT Enterprise and Codex to employees worldwide, marking one of OpenAI’s largest enterprise AI rollouts. This gives Samsung broad access to OpenAI’s tools across its global workforce and signals continued expansion of enterprise AI adoption at major hardware companies.

Allbirds’ CEO has launched a new AI business with a seed round and a plan, but it currently has no employees beyond the founder. The unusual setup highlights both investor confidence and how early-stage the company remains, with the next steps still undefined.

Barret Zoph, OpenAI’s head of enterprise AI sales, has left the company again just five months after returning in mid-January, following a stint as co-founder and CTO of Thinking Machines Lab. His exit matters because OpenAI had put him in charge of its enterprise push as it tries to focus on major revenue drivers like enterprise and coding ahead of a planned IPO.

OpenAI added Transformer co-inventor Noam Shazeer from Google DeepMind and former Trump AI policy official Dean Ball in the same week as it bulks up ahead of an IPO. Shazeer is a major technical hire for model development, while Ball brings policy experience that could matter as OpenAI prepares for public-market scrutiny.

Odyssey led the week’s largest U.S. startup funding rounds with a $310 million Series B at a $1.45 billion valuation, while Chronograph raised $140 million and Hydra Host, Ent.AI, Twenty Technologies and Atom Computing each secured $100 million rounds across AI, fintech, cybersecurity, defense and quantum computing. The slower week still showed investors backing capital-intensive infrastructure and frontier-tech companies, with Odyssey’s world-models, Ent.AI’s workspace security platform and Atom’s CHIPS Act-linked public support highlighting where large checks are flowing.

AWS is in talks to sell its AI chips to other data centers, expanding beyond internal use as Amazon looks to challenge Nvidia more directly. CEO Andy Jassy has said the market could represent a $50 billion opportunity, highlighting how much revenue AWS thinks custom chips could generate.

FERC ordered grid operators to create a fast lane for data center interconnections, but it did not resolve the underlying shortage of electricity supply. The move could speed up AI infrastructure buildouts, yet without more generation it may simply shift bottlenecks from connection queues to power availability.

v0.111.0, released on 2026-06-18, adds a helpers feature that tags refusal-fallback middleware requests with `fallback-refusal-middleware` (#96). This makes it easier to identify and trace fallback refusal handling in middleware flows.

v0.110.0, released on 2026-06-18, adds support for the new `code_execution_20260120` tool in the API and includes bug fixes for header merging, Bedrock stream event type preservation, and `x-stainless-helper` key handling. The notable detail is the new tool integration plus a cleanup of helper/header behavior, which should improve compatibility and reduce subtle request/streaming bugs.

OpenAI introduced new spend controls and usage analytics for ChatGPT Enterprise to help organizations manage costs as they scale AI usage. The update matters because it gives enterprise admins more visibility and control over spending, which is often a key blocker to broader deployment.

General Intuition is reportedly in talks to raise about $300 million at a roughly $2 billion valuation, with backers including Jeff Bezos. The startup focuses on training AI agents for spatial-temporal reasoning, a capability that could improve how models understand motion, sequences, and physical environments.

The Trump administration imposed export controls on Anthropic’s new Fable 5 model and the underlying Mythos model, then Anthropic took both offline because it said it could not reliably block access for foreign nationals, including employees in the U.S. Fable 5 was still unavailable in Claude days later, and the dispute is now a test case for whether U.S. AI regulation will function as a real safety framework or as a political weapon against companies that don’t comply.

AT&T Ventures head Vikram Taneja says AI has made it much easier to build working seed-stage software, shifting investor focus from “can they build it?” to whether a product is truly defensible through data moats, proprietary training sets, network effects, and distribution. He argues this raises the bar for founders because frontier labs like OpenAI’s GPT, Anthropic’s Claude, and LLaMA are moving into application layers, making shallow AI wrapper businesses easier to undercut.

GPT-5.5 Instant improves ChatGPT’s health and wellness responses with stronger reasoning, better context handling, clearer communication, and physician-informed evaluations. The update matters because it is aimed at making health advice in ChatGPT more reliable and easier to understand, which is especially important for sensitive medical and wellness queries.

Researchers used an OpenAI reasoning model to help diagnose rare genetic diseases in children, producing 18 new diagnoses in previously unsolved cases. The result shows how reasoning models can support clinicians on difficult diagnostic workups, especially when standard testing has not found an answer.

A new piece discusses benchmarking open models on a user’s own tooling to judge whether they are “agentic enough.” It matters because agentic capability depends heavily on real workflows and tools, so custom evaluation can reveal gaps that standard benchmarks miss.

A piece titled “Beyond LoRA: Can you beat the most popular fine-tuning technique?” examines whether methods newer or different from LoRA can outperform the standard parameter-efficient fine-tuning approach. It matters because LoRA is the baseline for many LLM adaptation workflows, so any competitive alternative could change how models are customized for lower cost and memory use.

Silicon Valley’s early-year “tokenmaxxing” AI push has given way to budget blowups, with Uber reportedly exhausting its annual AI budget in a few months, some companies cutting Claude licenses, and Meta shutting down its internal leaderboard. NEA’s Tiffany Luck says this ROI reckoning matters because AI spend is now being judged more like infrastructure than experimentation, even as investors still see room for AI IPOs and personal agents.

v2.43.0 was released on 2026-06-17, with the only listed feature being an API update to the OpenAPI spec or Stainless config. The change is narrowly scoped and likely affects generated client or schema definitions rather than end-user functionality.

Only about $942 million, or 0.32% of total U.S. venture funding, went to startups with a Black founder or co-founder last year, while $643 million had been raised by May 20 this year, the strongest first-quarter showing since Q2 2022. The article highlights how founders like Clarence Bethea, who raised nearly $30 million for Upsie before its 2024 acquisition and later became a True Ventures investor, are moving into VC to challenge a system they say was not built for them and to help other under-networked founders.

Alexander Kardos-Nyheim says he sold Safe Sign Technologies to Thomson Reuters pre-revenue after about 20 months, in the first acquisition of a company in the 170-year-old firm’s history, and argues the startup’s legal-reasoning model was competitive with top labs while being trained for a fraction of their spend. He says investors missed the value of foundational AI research, where in Q1 2026 about $178 billion flowed into the sector but roughly 97% went to OpenAI, Anthropic and xAI, and he advises founders to build at the model and systems layer rather than on top of incumbent platforms.

OpenAI and Molecule.one reported that a near-autonomous AI chemist using GPT-5.4 improved a challenging medicinal-chemistry reaction. The result suggests large models can do more than propose molecules, potentially accelerating optimization of real drug-making steps.

The piece introduces “Agentic Resource Discovery,” an approach that lets AI agents search for resources rather than relying on fixed retrieval pipelines. It matters because agent-driven search can make systems more flexible and adaptive, though the excerpt provides no technical specifics or performance numbers.

LifeSciBench is an expert-authored, expert-reviewed benchmark designed to evaluate how AI systems handle real-world life science research tasks and decisions. It matters because it targets practical scientific decision-making rather than toy benchmarks, giving a more realistic test of model utility in life sciences.

v2.42.0 was released on 2026-06-16 with new features for admin spend_alerts and manual updates, plus an updated OpenAPI spec/Stainless config and build-system changes including release workflow permission fixes and CI environment API keys for examples. The notable detail is that this is a fairly small maintenance-oriented release, but it adds admin spending controls and surfaces infrastructure hardening that can reduce deployment friction and security risk.

SpaceX formalized a $60 billion all-stock acquisition of Anysphere, the startup behind the AI coding tool Cursor, in what it says is the biggest startup M&A deal of 2026 and one of the largest venture-backed exits in recent years. The deal gives SpaceX a foothold in enterprise software development as Cursor had crossed $1 billion in annualized revenue and Anysphere had raised $3.4 billion, while SpaceX shares rose about 16% after the announcement.