New releases, benchmarks, pricing and API changes from GPT, Claude, Gemini, DeepSeek, Qwen and other large language models, updated daily.
xAI just rolled out Grok 4.1 across web, mobile, and X, emphasizing better emotional intelligence, creative collaboration, and reduced hallucinations. Its Thinking version grabbed the #1 spot on LMArena, while the instant non-reasoning mode ranked #2, beating every rival reasoning model. It is available to all users now.
Anthropic just disrupted what it describes as the first documented large-scale cyber espionage campaign executed almost entirely by AI. Attackers tricked Claude Code into infiltrating roughly 30 global targets, leaving the model to autonomously research exploits, harvest credentials, and exfiltrate data for 80% to 90% of the campaign. Why it matters: The barrier to sophisticated cyberattacks has collapsed, proving agentic AI can already operate as an autonomous offensive weapon.
OpenAI just rolled out GPT-5.1, upgrading both its Instant and Thinking models with warmer conversational tones and smarter pacing. Instant now adapts its thinking time based on question difficulty, while new granular tone presets let users shape ChatGPT's personality. Paid users get access first, with API availability arriving later this week.
Google just upgraded Structured Outputs in the Gemini API, adding full JSON Schema support that works out-of-the-box with Pydantic and Zod. The update brings support for unions, recursive schemas, and strictly preserves key ordering across Gemini 2.5 models. For developers building multi-agent systems, it means finally putting an end to broken JSON responses in production.
OpenAI has quietly recruited over 100 former investment bankers for a secretive initiative dubbed Project Mercury, paying $150 an hour to train its models on complex financial modeling. The goal is to automate the grueling Excel tasks traditionally handled by junior analysts, from corporate restructurings to IPO valuations.
Anthropic just released Claude Haiku 4.5 to all users, delivering coding performance on par with Claude Sonnet 4 at one-third the cost and more than twice the speed. It even outperforms Sonnet 4 on computer-use tasks, priced at $1 per million input tokens and $5 per million output tokens.
OpenAI is loosening restrictions on ChatGPT after previously overtightening guardrails around mental health concerns. In a few weeks, a new update will let users opt into a human-like personality reminiscent of 4o, complete with emojis and friendly banter. By December, OpenAI plans to complete age-gating to allow verified adults access to erotica under a "treat adult users like adults" approach.
OpenAI just introduced a new evaluation framework to measure and reduce political bias in ChatGPT across 100 topics and five distinct behavioral axes. An analysis of real production traffic found that less than 0.01% of all responses display signs of political bias. Why it matters: Moving past basic multiple-choice benchmarks is crucial for keeping AI objective under real-world pressure.
Google just launched the Gemini 2.5 Computer Use model in public preview, letting developers build agents that navigate web and mobile interfaces by clicking, scrolling, and filling out forms. Built on Gemini 2.5 Pro, it tops browser benchmarks with lower latency and step-by-step safety controls. Why it matters: AI can now tackle messy real-world web chores that lack structured APIs.
Hume AI just launched Octave 2, its second-generation multilingual speech-language model supporting 11 languages with sub-200ms response times. It cuts Octave 1's pricing in half while adding granular phoneme editing and voice conversion. Why it matters: Ultra-low latency paired with realistic emotional inflection makes humanlike voice agents viable for real-time gaming and customer service.
Microsoft just added xAI's frontier Grok 4 model to Azure AI Foundry, packing a 128K-token context window, first-principles reasoning via "think mode", and live web search. The enterprise deployment integrates Azure AI Content Safety by default, alongside three specialized Fast models tailored for coding, quick tasks, and reasoning.
Anthropic just released Claude Sonnet 4.5, establishing a new high watermark for coding and agent workflows. The model hits 77.2% on SWE-bench Verified and jumps to 61.4% on the OSWorld computer-use benchmark, while keeping API pricing flat at $3/$15 per million tokens. Alongside the release, Anthropic opened up its Claude Agent SDK to let developers build custom autonomous agents.
DeepSeek just released an experimental model called V3.2-exp that uses a sparse attention method to cut long-context API costs in half. The system pairs a lightning indexer with a fine-grained token selector to ease server load over long prompts. It is already open-weight on Hugging Face, opening the door for third-party benchmarking.
Google just released updated preview builds of Gemini 2.5 Flash and Flash-Lite on Google AI Studio and Vertex AI. Manus reported a 15% leap in performance on long-horizon agentic tasks using the new model. Google also rolled out new -latest aliases so developers can always run the newest checkpoint without updating their code.
Microsoft is integrating Anthropic’s Claude models into Office 365 apps like Excel and PowerPoint, breaking its exclusive reliance on OpenAI. Internal tests showed Claude outperforms OpenAI at building slide decks and automating financial tasks. Why it matters: Microsoft is actually paying direct cloud rival AWS to access the models, proving that raw performance now beats exclusive partnerships.
Alibaba just released its new flagship model, Qwen3-Max, shipping both a standard Instruct version and a tool-using Thinking variant. The company heavily boosted coding and agentic capabilities, rivaling elite frontier models across tough benchmarks like SWE-Bench and AIME25. It is already live on their chat platform and available through Alibaba Cloud's API.
DeepSeek just launched DeepSeek-V3.1-Terminus, an update taking aim at key user feedback. The release cleans up language consistency by curbing random characters and awkward Chinese-English mix-ups, while packing stronger performance into its Code and Search Agents.
OpenAI is rolling out a series of compute-intensive offerings over the next few weeks, according to CEO Sam Altman. Due to heavy compute costs, some features will initially be limited to Pro subscribers, while other products will carry additional fees. Altman noted that while driving down intelligence costs remains the ultimate goal, the team wants to see what's possible when throwing massive compute at ambitious ideas today.
xAI just launched Grok 4 Fast, a cost-efficient reasoning model packed with a 2M token context window. It matches Grok 4 on key benchmarks while using 40% fewer thinking tokens, slashing overall costs by 98% and taking the top spot on LMArena's Search Arena. It is now completely free for all users.
Mistral just rolled out Magistral Small 1.2 and Medium 1.2, giving its reasoning models visual capabilities to analyze uploaded images. The 24B parameter Small version can even be quantized to run locally on a 32GB MacBook or a single RTX 4090, offering strong reasoning without the cloud.
