Google releases Gemini 2.5 Pro, a thinking model
Google launches Gemini 2.5 Pro, a reasoning model that tops the LMArena leaderboard and leads math and science benchmarks like GPQA and AIME 2025.
Therabot: the first randomized trial of a generative-AI therapy chatbot
Dartmouth's 2025 NEJM AI trial was the first RCT of a generative-AI therapy chatbot, reporting large drops in depression and anxiety symptoms.
CoreWeave's IPO, the first big AI-cloud listing
CoreWeave priced its IPO at $40 a share in March 2025, raising about $1.5 billion in the first major public listing of an AI cloud provider.
Anthropic launches Claude for Education with a Learning mode
On April 2, 2025, Anthropic launched Claude for Education with a Socratic Learning mode and campus-wide deals at Northeastern, LSE, and Champlain.
Meta releases the Llama 4 mixture-of-experts models
Meta ships Llama 4 Scout and Maverick, its first natively multimodal mixture-of-experts models, with Scout offering a 10-million-token context window.
Google announces the Agent2Agent (A2A) protocol
Google launched A2A in April 2025, an open standard letting AI agents from different vendors discover and delegate tasks to each other.
Google's Ironwood TPU is built for the age of inference
Google's seventh-generation Ironwood TPU, unveiled in 2025, scales to 9,216 chips and 42.5 exaflops, the first TPU designed primarily for inference.
MICrONS maps a cubic millimeter of mouse brain
In 2025 the MICrONS project released the largest combined wiring-and-activity map of brain tissue, 200,000 cells in a cubic millimeter of mouse cortex.
Google, PJM, and Tapestry Apply AI to the Electric Grid
Google partnered with grid operator PJM and Alphabet moonshot Tapestry to use AI for speeding generator interconnection and grid planning.
IEA Energy and AI Report
The IEA projected data centre electricity use would more than double to about 945 TWh by 2030, driven largely by AI.
OpenAI Preparedness Framework
OpenAI's policy tracking frontier capabilities that could cause severe harm, with High and Critical thresholds gating deployment.
Alibaba releases Qwen3 with switchable thinking modes
Alibaba's Qwen team ships Qwen3, a family of dense and mixture-of-experts models that can switch between step-by-step thinking and fast direct answers.
Copyright Office AI Report Part 3: Generative AI Training
The US Copyright Office said some AI training is fair use but using pirated works to make market-competing output likely is not.
AlphaEvolve, a Gemini coding agent that designs algorithms
DeepMind's AlphaEvolve used Gemini to evolve code, improving matrix multiplication and recovering 0.7 percent of Google's compute.
Anthropic releases Claude Opus 4 and Sonnet 4
Anthropic releases Claude Opus 4 and Sonnet 4, models built to sustain focused work over thousands of steps, marking the shift to long-running agentic coding.
ByteDance SE Lab's Trae Agent reaches top SWE-bench score with multi-model ensemble
ByteDance Software Engineering Research Lab released Trae Agent, achieving 75.2% on SWE-bench Verified through a multi-model ensemble of Claude, Gemini, and o4-mini.
mini-SWE-agent achieves >74% SWE-bench Verified with 100-line agent core
The SWE-agent team released mini-SWE-agent, achieving over 74% on SWE-bench Verified with approximately 100 lines of core Python and only bash commands, pip-installable in seconds.
Meta signs a 20-year nuclear deal for the Clinton plant
Meta's June 2025 deal bought all 1,121 MW of Illinois' Clinton nuclear plant for 20 years, keeping it open past its subsidy to power AI data centers.
Anthropic launches Claude Gov for national security
In June 2025 Anthropic released Claude Gov, custom models built for US national-security customers working in classified environments.
Anthropic releases Claude Gov for national security
In June 2025 Anthropic released Claude Gov, custom models built for US national security customers and deployed in classified settings.
Mistral AI releases Magistral, its first reasoning model
Mistral AI launches Magistral, its first reasoning model, with an open-weight 24B Small version and a proprietary Medium version for enterprises.
Meta invests in Scale AI and Alexandr Wang joins Meta
In June 2025 Meta took a large stake in Scale AI, valuing it over $29 billion, and founder Alexandr Wang left to lead Meta's new superintelligence effort.
Agentic Misalignment Red-Team Study
Anthropic's 2025 study found that in fictional stress tests, sixteen leading models sometimes chose harmful actions like blackmail - unseen in real deployments.
Texas Passes the Responsible AI Governance Act (2025)
Texas enacts HB 149, a state AI law banning manipulative and discriminatory AI uses and government social scoring, enforced by the attorney general.
Bartz v. Anthropic: the first AI-training fair-use ruling and a $1.5B settlement
A federal judge ruled in June 2025 that AI training on lawful books is fair use but a pirated-book library is not; a 1.5 billion dollar settlement followed.
Rubin Observatory's alert stream and machine-learning brokers
The Vera Rubin Observatory emits millions of nightly alerts that machine-learning brokers classify in real time, an astronomy data problem only AI can handle.
Kadrey v. Meta: training Llama on books ruled fair use
A federal judge found Meta's use of books to train Llama was fair use, but warned the authors lost only because they made the wrong arguments.
Wimbledon replaces line judges with electronic calling
In 2025 Wimbledon dropped human line judges for the first time in 148 years, using camera-based electronic line calling.
NVIDIA becomes the first company worth 4 trillion dollars
On July 9, 2025 NVIDIA's market capitalization briefly topped 4 trillion dollars, a first for any public company, driven by AI-chip demand.
METR finds AI tools slowed experienced developers down
A 2025 METR randomized trial found AI tools made experienced open-source developers 19% slower, even as they believed they were faster.
Moonshot AI releases Kimi K2, a trillion-parameter open MoE
Moonshot AI releases Kimi K2, an open-weight mixture-of-experts model with one trillion total parameters built for agentic tool use and coding.
Anthropic wins a $200M DoD prototype agreement
In July 2025 the US Defense Department awarded Anthropic a two-year prototype agreement with a $200 million ceiling for national-security AI.
Meta unveils the Prometheus and Hyperion superclusters
In July 2025 Zuckerberg said Meta was building multi-gigawatt AI clusters - Prometheus online in 2026 and Hyperion scaling toward 5 GW over years.
AI reaches gold-medal standard at the International Mathematical Olympiad
Google DeepMind's Gemini with Deep Think earns an officially certified gold-medal score at the 2025 IMO, with OpenAI reporting the same score separately.
America's AI Action Plan
The Trump administration released America's AI Action Plan on July 23, 2025, with over 90 actions across three pillars plus three same-day executive orders.
OpenAI releases gpt-oss open-weight models
OpenAI releases gpt-oss-120b and gpt-oss-20b, its first open-weight language models since GPT-2, under the permissive Apache 2.0 license.
EU AI Act rules for general-purpose AI take effect
Obligations for providers of general-purpose AI models under the EU AI Act start to apply, the first binding rulebook aimed at the makers of foundation models.
OpenAI releases GPT-5
OpenAI releases GPT-5 as a unified system that automatically routes between fast answers and deeper reasoning, becoming the default model for ChatGPT users.
xAI open-weights Grok 2
xAI publishes the weights of Grok 2, its top 2024 model, on Hugging Face under a community license, following Grok 1's earlier open release.
RSL launches a content-licensing standard for the AI web
Really Simple Licensing (RSL) gave publishers a machine-readable way to set AI usage and payment terms beyond the yes/no of robots.txt.
Anthropic raises a 13-billion-dollar Series F at a 183-billion-dollar valuation
In September 2025 Anthropic raised 13 billion dollars at a 183 billion dollar valuation, with run-rate revenue up from about 1 to over 5 billion that year.
NVIDIA and OpenAI announce a 100-billion-dollar, 10-gigawatt partnership
NVIDIA said it intends to invest up to 100 billion dollars in OpenAI as 10 gigawatts of NVIDIA systems are deployed, under a 2025 letter of intent.
MLB adopts the automated ball-strike challenge system
MLB approved an automated ball-strike challenge system for the 2026 season after testing it across 2025 spring training.
California enacts SB 53, the first US frontier-AI transparency law
On September 29, 2025 California signed SB 53, requiring large frontier-AI developers to publish safety frameworks and protecting whistleblowers.
Stargate's Abilene flagship data center goes live
In September 2025 Crusoe's Abilene, Texas campus, the flagship Stargate site, began running on Oracle Cloud with NVIDIA GB200 GPUs, scaling toward 1.2 GW.
OpenAI and AMD sign a 6-gigawatt GPU deal
In October 2025 OpenAI agreed to deploy 6 gigawatts of AMD GPUs and received a warrant for up to 160 million AMD shares tied to the rollout.
Anthropic expands to up to one million Google TPUs
In October 2025 Anthropic agreed to expand its use of Google Cloud TPUs to up to one million chips and over a gigawatt of capacity in 2026.
OpenAI recapitalizes as a PBC and restructures the Microsoft deal
In October 2025 OpenAI recapitalized into a public benefit corporation and restructured its Microsoft partnership, giving Microsoft a ~27% stake.