Historical news archive. Items here are older than the 24-hour Latest News window. This page is for reference; the live AI news feed is at /news/.
When you generate release notes for a new release, the generated notes include a list of pull requests merged since the last release. For more information about automatically generating release...
An open-source release from github.blog relevant to AI implementation and tooling.
Researchers used an OpenAI reasoning model to help diagnose rare diseases, identifying 18 new diagnoses in previously unsolved cases.
A model-relevant release from openai.com with potential impact for AI builders.
arXiv:2606.18543v1 Announce Type: new Abstract: Language model agents are becoming proficient executors at isolated, short-horizon tasks such as software engineering and customer service. Yet real-world challenges require a combination of sophisticated skills that remain largely untested in agents: (1) navigating long
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18557v1 Announce Type: new Abstract: A rule-based logic solver resolves every instance in our benchmark in under 50 microseconds with 100% accuracy; the best frontier language model reaches 65% at best and drops to 23.5% under rendering-robust evaluation (worst case over four surface renderings). We introduc
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.19116v1 Announce Type: new Abstract: The World Wide Web was built on an assumption held for three decades: the primary consumer of web content is a human being. This permeates every layer; its access model presumes human visitors, its economics rest on human attention, and its content targets human perceptio
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.19118v1 Announce Type: new Abstract: Electricity markets are inherently complex systems characterised by strong nonlinearities, high-dimensional interactions, and increasing interdependence across regions. While deep neural networks (DNNs) have demonstrated strong predictive capabilities for electricity pric
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18293v1 Announce Type: cross Abstract: Thanks to rapid developments in generative AI, we are in the midst of a paradigm shift that may change how we interact with computers forever. We have observed a growth in the use of natural language prompts to build applications and coding infrastructures without under
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18356v1 Announce Type: cross Abstract: Tool-using language-model agents introduce security failures that go beyond unsafe text: they can disclose protected objects, write persistent memory, send messages, modify databases, or trigger harmful code and tool effects. Existing evaluations often collapse these st
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18379v1 Announce Type: cross Abstract: Graph-based retrieval at billion-node scale requires jointly solving three tightly coupled problems -- graph construction, representation learning, and real-time serving -- yet existing work addresses each in isolation. We present RankGraph-2, a framework deployed at Me
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18485v1 Announce Type: cross Abstract: Neural Text-to-Speech (TTS) systems achieve remarkable quality on short utterances but long-form speech generation shows prosodic drift, speaker inconsistencies and sentence boundary artifacts. Existing approaches either compress sequences, increase context length or na
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18661v1 Announce Type: cross Abstract: Intelligent landslide hazard interpretation is critical for disaster prevention, yet current paradigms struggle to simultaneously extract visual features and high-level geoscientific semantics, while general-purpose vision-language models (VLMs) suffer from perceptual l
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18717v1 Announce Type: cross Abstract: Turkish is agglutinative: meaning is carried by morphemes, yet the subword tokenizers that drive modern language models split words by corpus statistics, fragmenting semantically loaded suffixes and -- in the case of WordPiece and rule-based analyzers -- failing to deco
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18837v1 Announce Type: cross Abstract: Large Language Model (LLM)-based automatic Multi-Agent Systems (MAS) generation has become a crucial frontier for tackling complex tasks. However, existing methods face a dilemma between model capability and experience retention.
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18864v1 Announce Type: cross Abstract: This paper studies how to scale learning-based automatic emergency braking (AEB) with massive unlabeled fleet data under production constraints. Our approach is based on meta-feedback semi-supervised learning (MF-SSL), where a teacher generates pseudo labels for unlabel
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.19135v1 Announce Type: cross Abstract: As large language models (LLMs) advance and multi-agent systems aim to overcome the limits of standalone agents, robust communication protocols are becoming essential infrastructure for distributed agent networks. Nonetheless, the fragmented protocol landscape presents
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.19259v1 Announce Type: cross Abstract: Text-rich images often contain privacy-sensitive, transactional, or decision-relevant information. As recent multimodal image generation models become increasingly capable of synthesizing realistic textual content and structured visual designs, detecting AI-generated te
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.19328v1 Announce Type: cross Abstract: Preference-based RL provides an approach to learning reward models from pairwise comparisons of behaviors, bypassing the need for explicit reward design. However, existing methods typically rely on passive data collection and suffer from poor sample efficiency, especial
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2508.21720v3 Announce Type: replace Abstract: Automating scientific poster generation requires hierarchical document understanding and coherent content-layout planning. Existing methods often rely on flat summarization or optimize content and layout separately.
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2512.04144v2 Announce Type: replace Abstract: Targeted interventions on language models, such as unlearning or model editing, aim to modify specific information, but their effects often propagate to related, unintended areas (e.g., removing virology content may degrade performance on allergies); these side-effect
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.11918v2 Announce Type: replace Abstract: Current Large Reasoning Models (LRMs) exhibit remarkable general capabilities but significantly underperform in spatial reasoning tasks. Existing approaches treat this gap as a knowledge deficit, relying on supervised fine-tuning (SFT) to ingest labeled spatial data f
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2303.18031v2 Announce Type: replace-cross Abstract: In real-world applications, a machine learning model is required to handle an open-set recognition (OSR), where unknown classes appear during the inference, in addition to a domain shift, where the data distribution differs between the training and inference pha
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2502.07531v5 Announce Type: replace-cross Abstract: Controllable image-to-video (I2V) generation transforms a reference image into a coherent video guided by user-specified control signals. While precise control over camera motion, object motion, and lighting is essential for high-fidelity creation, existing meth
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2505.16057v2 Announce Type: replace-cross Abstract: AI-Generated (AIG) content has become increasingly widespread by recent advances in generative models and the easy-to-use tools that have significantly lowered the technical barriers for producing highly realistic audio, images, and videos through simple natural
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2506.14126v2 Announce Type: replace-cross Abstract: Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized datasets. This has led to a proliferation of expert models and adapters, often shared via platforms like HuggingFace and AdapterH
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2510.04120v2 Announce Type: replace-cross Abstract: Large language models (LLMs) achieve strong performance on metaphor detection and interpretation tasks, yet it remains unclear what such behavioral success reveals about metaphor processing. We present a diagnostic analysis that examines the limits of behavioral
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2602.00161v2 Announce Type: replace-cross Abstract: In this paper, we formulate the compression of large language models (LLMs) by optimally deleting transformer blocks (``block removal'') as a constrained binary optimization (CBO) problem that can be mapped to a physical system (Ising glass), whose energies are
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2604.04089v4 Announce Type: replace-cross Abstract: Large language models can write scientific code, but direct paper-to-program translation remains fragile when correctness depends on tacit conventions rather than explicit equations. We frame this as a \textbf{knowledge-externalization} problem: index choices, g
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2604.06367v2 Announce Type: replace-cross Abstract: Web agents automate browser tasks, ranging from simple form completion to complex workflows like ordering groceries. While current benchmarks evaluate general-purpose performance~(e.g., WebArena) or safety against malicious actions~(e.g., SafeArena), no existing
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2605.12713v3 Announce Type: replace-cross Abstract: In the field of quantum reservoir computing (QRC), many different computational models and architectures have been proposed. From these models, we identify feedback-based models -- which use a feedback mechanism to re-embed classical measurements from the QRC
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2605.21028v4 Announce Type: replace-cross Abstract: Autoregressive long video generation often adopts bounded-memory streaming for efficiency, typically combining local windows for short-term continuity with static early-frame sinks as long-range anchors. However, this fixed allocation keeps early frames cached e
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.02045v2 Announce Type: replace-cross Abstract: Artificial intelligence provides a practical framework for crop damage assessment from imagery data, supporting early decision-making in agricultural management. In peach orchards, climate change increases abiotic stress and biotic pressures, including pests and
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.02800v3 Announce Type: replace-cross Abstract: We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configuration
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.07150v3 Announce Type: replace-cross Abstract: Agent-interoperability protocols such as A2A and MCP standardize what agents say to one another but assume address-based transport. Whether over HTTP(S) or a content-protecting binding such as MLS-based SLIM, these transports protect message content yet leave th
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.12629v2 Announce Type: replace-cross Abstract: We show the standard basis of transformer hidden states already provides a training-free, architecture-general feature basis. Individual dimensions encode semantic content via their signs (+/-1) and confidence via their magnitudes, acting as independent binary r
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.14202v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have advanced Automatic Heuristic Design (AHD) by enabling heuristic generation through reasoning and code synthesis. Existing LLM-based AHD architectures mainly follow two paradigms: Natural Evolution, which uses crossover and mutat
A research signal from arxiv.org likely to affect model direction and evaluation choices.
Explore repository content from your terminal without cloning by using two new commands: gh repo read-file and gh repo read-dir. What you can do Use gh repo read-file to read...
An open-source release from github.blog relevant to AI implementation and tooling.
Since our last pattern update, we’ve expanded secret scanning’s detection coverage with new partners, more patterns blocked by push protection by default, additional validity checks, and richer metadata for leaked...
An open-source release from github.blog relevant to AI implementation and tooling.
We’re adding our first governance capability to the enterprise-managed settings configuration. Enterprise administrators can now set disableBypassPermissionsMode to "disable" in the enterprise-managed settings.json to prevent GitHub Copilot CLI and VS...
An open-source release from github.blog relevant to AI implementation and tooling.
GitHub Copilot auto model selection is now generally available in Copilot Chat on github.com and the GitHub mobile app for all Copilot plans. With auto, Copilot chooses a model on...
An open-source release from github.blog relevant to AI implementation and tooling.
Maintainers of open source repositories are dealing with an ever-growing volume of pull requests, including repeated low-quality or drive-by contributions that can slow triage and overwhelm review queues. To help...
An open-source release from github.blog relevant to AI implementation and tooling.
GitHub agent finder is now available. Instead of hand-wiring which MCP servers, skills, canvases, agents, and tools each agent should use-and filling your context window in the process-GitHub Copilot can...
An open-source release from github.blog relevant to AI implementation and tooling.
We’re reopening sign-ups for GitHub Copilot Student, Pro, Pro+, and Max plans gradually over the next couple of weeks for new subscribers. If you’ve been waiting to get started with...
An open-source release from github.blog relevant to AI implementation and tooling.
The GitHub Copilot app is now generally available for macOS, Windows, and Linux. It’s the desktop home for agent-driven development, built natively on GitHub.
An open-source release from github.blog relevant to AI implementation and tooling.
OpenAI and Molecule.one show how a near-autonomous AI chemist using GPT-5.4 improved a key drug-making reaction, advancing medicinal chemistry research.
A model-relevant release from openai.com with potential impact for AI builders.
arXiv:2606.17637v1 Announce Type: new Abstract: Building Management Systems (BMS) are essential for optimizing energy efficiency and operational performance in modern buildings. However, the lack of standardization across BMS points from different manufacturers creates significant barriers to integration and data utili
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17645v1 Announce Type: new Abstract: Large language model (LLM) web agents are usually deployed as tool callers: each turn, the model reads a fresh page observation and emits one structured tool action. When every action is a low-level primitive, horizons grow quickly and so do policy-facing LLM completions
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17696v1 Announce Type: new Abstract: Parametric computer-aided design records both final geometry and the ordered construction history that determines how a part can be edited. Datasets for editable CAD research should therefore expose modeling operations, parameters, and feature dependencies together with v
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17727v1 Announce Type: new Abstract: Recent vision-language models (VLMs) have shown promising progress in generating webpages from visual inputs, yet existing evaluations mainly focus on short, single-screen, and largely static webpages. We introduce LongWebBench, a benchmark for evaluating long-horizon web
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18132v1 Announce Type: new Abstract: Meta-reinforcement learning enables fast adaptation by extracting shared structure from related tasks, but existing end-to-end methods often couple task inference with embodiment-specific control. This coupling can obscure non-parametric task semantics, reduce sample effi
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17059v1 Announce Type: cross Abstract: Prefix caching can reduce LLM inference latency by reusing KV caches across requests with shared prompts, but cluster-scale reuse is challenging because caches are partitioned across nodes. We propose a decentralized, prefix-cache-aware routing scheme for peer-to-peer L
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17077v1 Announce Type: cross Abstract: Proton dissociation constants (pKa) are critical for functional molecule discovery and molecular modeling. Building on iBonD, the largest experimental pKa database established, we and other researchers have developed several methods including machine-learning-based empi
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17081v1 Announce Type: cross Abstract: Disaggregated inference architectures physically separate prefill and decode phases onto distinct GPU pools, creating competing "agents" that share a fixed hardware budget. We provide, to our knowledge, the first formal game-theoretic analysis of this architecture, usin
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17090v1 Announce Type: cross Abstract: ANEForge is a Python package that programs the Apple Neural Engine (ANE), the fixed-function neural accelerator on every recent Apple device, directly and without CoreML. In production the engine is reachable only through CoreML, which treats it as a scheduling option:
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17379v1 Announce Type: cross Abstract: Accurate intraoperative liver registration is challenging due to substantial soft-tissue deformation yet sparse intraoperative measurements. Biomechanical models regularize this ill-posedness with prior knowledge but exhibit persistent prediction bias due to simplifying
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17646v1 Announce Type: cross Abstract: Saliency map visualizations explain image-based AI predictions by pointing to regions, but these are often unintuitive and semantically unclear, leaving an interpretability gap. We argue that AI explanations should be intuitive -- coherent to user knowledge, yet simple
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17660v1 Announce Type: cross Abstract: Fine-tuning large language models (LLMs) is compute-intensive and error-prone: model performance depends sensitively on data quality and hyperparameter choices, and na\"ive runs can even degrade model performance. This raises a practical question:can we predict fine-tun
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17702v1 Announce Type: cross Abstract: Characterising the tumour microenvironment (TME) from routine H&E-stained histology images requires simultaneous cell segmentation, feature extraction, and interpretable clinical reporting. We present SEGTME-UNI2, a unified framework addressing these requirements.
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17924v1 Announce Type: cross Abstract: Current Vision-Language-Action (VLA) models face a trade-off between efficient action generation and explicit deliberation. Directly decoding actions from vision-language backbone representations enables low-latency control, whereas explicit reasoning through textual ch
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.17989v1 Announce Type: cross Abstract: Multi-contrast magnetic resonance imaging (MRI) provides complementary information for clinical diagnosis. However, acquiring all MRI sequences is often time-consuming and costly.
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18120v1 Announce Type: cross Abstract: Large language model applications build prompts from templates, and Handlebars is a widely used templating engine and the default prompt-template format in Microsoft Semantic Kernel. Its double-brace {{x}} expression HTML-escapes the interpolated value and is documented
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18135v1 Announce Type: cross Abstract: In this work, we introduce the Certus Caliber Classification Gunshot Dataset (C3GD), a publicly accessible data set developed for the analysis of firearm muzzle blast sounds. The dataset aims to provide a wide variety of firearms, calibers, cartridges, microphones, and
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18156v1 Announce Type: cross Abstract: We present a novel framework for realistic and controllable 3D face re-aging which produces highly detailed, identity-preserving results. Existing 3D editing methods, while effective for coarse semantic changes, are not well suited for re-aging, as even small inconsiste
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18168v1 Announce Type: cross Abstract: Software practitioners increasingly use AI coding agents that generate test code alongside production code in open source pull requests (PRs). Recent studies report more than 932,000 agent-authored PRs across more than 116,000 repositories, yet whether their test files
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18193v1 Announce Type: cross Abstract: We evaluate the adversarial robustness of two frontier large language models (LLMs) developed by Anthropic, Fable 5 and Opus 4.8, against four families of automated jailbreak attack across 7 826 harmful intents spanning a ten-category harm taxonomy. Using the HackAgent
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2606.18203v1 Announce Type: cross Abstract: The LLM-empowered personal health agents with user health (sensor) metrics have offered a promising pathway to alleviate global disparities in healthcare access. However, large-scale clinical deployment remains constrained by an open-ended evaluation bottleneck: physici
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2605.30036v2 Announce Type: replace Abstract: Large Language Models (LLMs) demonstrate a remarkable capacity to adopt different personas and roles; however, it remains unclear whether they can manifest behavior that adheres to a coherent, human-like value structure. In this work, we draw on established psychologi
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2510.01359v2 Announce Type: replace-cross Abstract: Code-capable large language model (LLM) agents are embedded in software engineering workflows where they can read, write, and execute code, raising "jailbreak" stakes beyond text-only settings. Prior evaluations emphasize refusal or harmful-text detection, leavi
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2510.21583v3 Announce Type: replace-cross Abstract: Recent Progress in post-training flow matching for text-to-image (T2I) generation with Group Relative Policy Optimization (GRPO) has demonstrated strong potential. However, it is hindered by a critical limitation: inaccurate advantage attribution.
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2601.16407v4 Announce Type: replace-cross Abstract: Large language models (LLMs) make next-token predictions based on clues present in their context, such as semantic descriptions and in-context examples. Yet, elucidating which prior tokens most strongly influence a given prediction remains challenging due to the
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2603.03485v3 Announce Type: replace-cross Abstract: Recent video diffusion models have achieved impressive capabilities as large-scale generative world models. However, these models often struggle with fine-grained physical consistency, exhibiting physically implausible dynamics over time.
A research signal from arxiv.org likely to affect model direction and evaluation choices.
arXiv:2605.29179v2 Announce Type: replace-cross Abstract: Metal-organic frameworks (MOFs) are excellent candidates for water harvesting due to their tunable pore environments, which can be precisely engineered to capture and release water in arid conditions. Integrating artificial intelligence (AI) into MOF discovery c
A research signal from arxiv.org likely to affect model direction and evaluation choices.
Introducing LifeSciBench, an expert-authored, expert-reviewed benchmark for evaluating how AI systems handle real-world life science research tasks and decisions.
A research signal from openai.com likely to affect model direction and evaluation choices.
Varonis Threat Labs disclosed a 3-stage chain (parameter-to-prompt injection, HTML-injection race, and SSRF via Bing) in M365 Copilot Enterprise Search that turns Copilot into a one-click data exfiltration tool against any user who clicks an attacker-crafted link. Microsoft assigned CVE-2026-26137 and began rolling out
SearchLeak is a working, one-click enterprise data exfiltration chain against the most-deployed AI assistant in the Microsoft 365 stack, and the same parameter-to-prompt pattern recurs across many enterprise RAG systems.
GitHub now runs CodeQL, GitHub Advisory Database dependency checks, and secret scanning on code produced by third-party coding agents (Claude, OpenAI Codex) before the PR finalizes. The validation is on by default, with no action required from repo owners.
Default-on security scanning for AI-generated PRs changes the security baseline of every GitHub repo that enables coding agents, and the same model is now an industry template other code hosts are likely to copy.
The US Commerce Department ordered Anthropic on June 12 to cut off Fable 5 and Mythos 5 for all foreign nationals; Anthropic disabled both models globally rather than try to segment users. In-person talks between Anthropic leadership and White House officials are ongoing.
Anthropic's global takedown of two flagship models sets a major precedent for how US AI vendors respond to export-control orders and exposes the practical limits of national-segmentation strategies for frontier models.
Effective June 15, Pro/Max/Team/Enterprise plans get a fixed monthly 'Agent SDK credit' pool ($20 Pro, $100 Max 5x, $200 Max 20x, $20 Team Standard, $100 Team Premium). When exhausted, SDK and headless 'claude -p' calls no longer silently count toward the standard message limit and instead degrade or require an upgrade
Anthropic is moving agent and headless usage onto a separate, predictable billing lane, which is a significant change for anyone running autonomous agents or CI on Claude and forces a re-read of every cost projection.
OpenAI released GPT-5, describing a unified system that routes between a fast default model and a deeper reasoning model, with a 256k context window, vision input, and improvements on coding, math, and multimodal benchmarks.
GPT-5 is the default model behind ChatGPT and the OpenAI API, so this release changes what hundreds of AI products and coding tools can do for hundreds of millions of users.
Anthropic released Claude Opus 4.1, an upgrade to Claude Opus 4 focused on agentic tasks, real-world coding, and reasoning, with a 200k context window and improved tool use.
Opus 4.1 is the strongest Anthropic coding model and is the default in Claude Code for many enterprise coding workflows, so a release here directly affects AI-assisted software quality.
Google DeepMind released Gemini 2.5 Pro Deep Think, a reasoning mode that uses parallel thinking to tackle hard math, coding, and multimodal problems, available to Google AI Ultra subscribers.
Deep Think is the first widely available consumer-facing parallel-reasoning model, and it sets a new bar on hard reasoning benchmarks; pricing and access shape how teams adopt it.
Hugging Face released SmolLM3, a 3B-parameter open-weight model with both thinking and non-thinking modes, an Apache-2.0 license, and full training-data recipe; it competes with much larger proprietary models on reasoning benchmarks.
A 3B model that rivals larger closed models in reasoning and ships fully open changes what local AI can do on a laptop, and it gives builders a real base for fine-tuning without GPU cluster costs.
Mistral released Magistral, a reasoning model family in Small (24B, Apache-2.0) and Medium (enterprise) variants, optimized for chain-of-thought across math, coding, and multi-step agent tasks.
Mistral's first openly licensed reasoning model is a clear signal that the open-weight model race is now competing with closed labs on chain-of-thought quality, not just base chat quality.
GitHub launched a Copilot coding agent that can pick up GitHub issues, open a pull request, and iterate on CI feedback, with humans able to review and request changes at every step.
The Copilot coding agent brings autonomous repo work into the world's largest code-hosting platform, setting a new baseline for what 'agentic' means in mainstream developer tooling.
Meta released Llama 4 Scout (17B active, 10M context) and Llama 4 Maverick (17B active, 400B total) as open-weight, mixture-of-experts models with native multimodality across text, image, and video.
Open-weight models with 10M-token context and native multimodal training are new territory for the open ecosystem; they make long-context document and video tasks feasible without API costs.
DeepSeek released R1, an MIT-licensed open-weight reasoning model with publicly documented training pipeline and strong benchmark results on math, coding, and scientific reasoning, alongside the smaller R1-Distill family.
DeepSeek-R1 is the first open-weight reasoning model to match closed frontier models on key benchmarks at a fraction of the cost, and it changed how enterprises and governments think about open AI stacks.
OpenAI introduced Sora, a text-to-video model that can generate up to 60-second clips with multi-shot scenes, persistent characters, and physics-aware motion, and rolled it out to ChatGPT and the OpenAI API in stages.
Sora is the first widely available consumer-facing text-to-video model at this fidelity, and it kicked off a wave of competing video-generation launches from Google, Meta, Runway, and others.
Anthropic released the Model Context Protocol (MCP), an open standard that lets AI assistants connect to data sources and tools through a single interface, and donated it to a new open-governance foundation.
MCP is becoming the de facto standard for connecting agents to tools and data; supporting it has become table stakes for any serious coding agent, IDE, or enterprise AI platform.
Anthropic released a beta of Computer Use, a Claude capability that lets the model see and operate a real desktop by taking screenshots, moving the cursor, and clicking, with safety guidance for high-impact actions.
Computer Use turned Claude into a general-purpose browser-and-desktop agent and kicked off the wave of GUI agents that now ship in browsers, IDEs, and enterprise automation products.
OpenAI released the Realtime API, with speech-to-speech models that handle interruptions, function calling, and tone detection, enabling production voice agents in a single API call.
Production voice agents previously required stitching together STT, LLM, and TTS; the Realtime API made it possible to ship conversational voice products in days rather than months.
OpenAI released Structured Outputs, which guarantees model responses match a developer-supplied JSON schema, and made the feature available across GPT-4o and its smaller models at no extra cost.
Reliable JSON-schema compliance is the foundation for production agent systems, tool calls, and database writes, and Structured Outputs removes the biggest source of brittleness in production LLM code.
Anthropic released Claude 3.5 Sonnet, with state-of-the-art results on coding, reasoning, and vision benchmarks at one-fifth the cost of Claude 3 Opus, plus a 200k context window and an upgraded Artifacts workspace.
Claude 3.5 Sonnet set a new cost-to-capability ratio for frontier models and became the default for many coding-agent products, so its release reshaped pricing across the API market.