Nous Research releases Hermes 4
Hermes 4 combined open-weight reasoning with general instruction following across several model sizes.
Artificial Intelligence Blogs
Short posts on models, agents, APIs, and the companies building them
Latest posts
355 entriesAugust 2025
Detecting and Countering Malicious Uses of Claude
Anthropic posted “Detecting and Countering Malicious Uses of Claude.” A quick update for the record.
Claude Code and new admin controls for business plans
Anthropic posted “Claude Code and new admin controls for business plans.” A quick update for the record.
July 2025
OpenAI launches ChatGPT agent
OpenAI launched ChatGPT agent on July 17, 2025, allowing the system to operate tools and complete multi-step tasks.
June 2025
Claude artifacts can call Claude through an API
Shared artifacts could become AI-powered apps whose usage runs against each visitor’s Claude account.
OpenAI releases o3-pro in the API
A higher-compute o3 variant reached Responses and Batch while standard o3 pricing fell.
GPT-4.1 models gain preference fine-tuning
OpenAI added direct preference optimization fine-tuning for the GPT-4.1 family.
OpenAI releases the Agents SDK for TypeScript
The agent orchestration SDK expanded beyond Python to TypeScript.
May 2025
Ukraine reports GOGOL-M autonomous drone trial
Ukrainian developers announced a GOGOL-M trial in which a mothership released smaller strike drones that navigated toward assigned targets.
Anthropic API adds Files, MCP, and longer prompt caching
A Files API, managed MCP connector, and one-hour prompt cache arrived together.
Anthropic API adds code execution
Claude gained a sandboxed Python execution tool for analysis and visualizations.
Responses API adds background mode
Long-running responses could execute asynchronously with improved reliability.
Responses API adds Code Interpreter and image tools
OpenAI expanded Responses with code execution, image generation, and improved file search.
Responses API adds remote MCP servers
OpenAI let Responses connect directly to remote Model Context Protocol servers.
Google announces MCP support for the Gemini API
Gemini SDK tooling began supporting the Model Context Protocol for external tools and data.
Gemini API adds URL context
Developers could pass links for Gemini to use as context in a request.
Gemini API adds asynchronous function calling
Long-running tools could return results later without blocking the active model session.
Gemini API previews native audio
Google previewed native audio output and a more expressive Live API at I/O.
Anthropic adds web search to the API
Claude API calls could retrieve current web information with citations.
OpenAI expands API data residency in Asia
Eligible customers gained at-rest data residency options in Japan, India, Singapore, and South Korea.
April 2025
OpenAI adds stronger API budget controls
Enhanced budget alerts and auto-recharge limits gave developers more control over spend.
OpenAI releases gpt-image-1 in the API
The image model behind ChatGPT became available for generation and editing through the API.
o3 and o4-mini arrive in the API
OpenAI released two reasoning models with stronger tool use and visual reasoning.
OpenAI releases the GPT-4.1 API family
GPT-4.1, mini, and nano emphasized coding, instruction following, and a one-million-token context window.
March 2025
OpenAI expands the Audio API
New speech-to-text and text-to-speech models targeted more capable voice agents.
o1-pro becomes available through the API
OpenAI released a higher-compute version of o1 through Responses and Batch.
Anthropic cuts token use for tool calls
Cache-aware rate limits and token-efficient tool use reduced cost and output tokens.
OpenAI releases the Agents SDK
The new SDK added orchestration, handoffs, guardrails, and tracing for agent workflows.
OpenAI launches the Responses API
A new API primitive combined model responses with built-in web, file, and computer tools.
Anthropic upgrades its developer Console
Anthropic refreshed the Console to shorten the path from prompt testing to production.
February 2025
GPT-4.5 enters the API
OpenAI made the GPT-4.5 research preview available to developers.
xAI releases Grok 3 beta
Grok 3 and Grok 3 mini added longer reasoning and were trained with substantially more compute on xAI's Colossus cluster.
OpenAI adds European data residency for API customers
Eligible API customers could choose Europe for data stored at rest.
OpenAI launches Deep Research
OpenAI launched Deep Research on February 2, 2025, enabling longer web research tasks with cited reports.
December 2024
Deliberative alignment: reasoning enables safer language models
OpenAI posted “Deliberative alignment: reasoning enables safer language models.” A quick update for the record.
Aligning on child safety principles
Anthropic posted “Aligning on child safety principles.” A quick update for the record.
Claude is now available in Brazil
Anthropic posted “Claude is now available in Brazil.” A quick update for the record.
A General Language Assistant as a Laboratory for Alignment
Anthropic posted “A General Language Assistant as a Laboratory for Alignment.” A quick update for the record.
Discovering Language Model Behaviors with Model-Written Evaluations
Anthropic posted “Discovering Language Model Behaviors with Model-Written Evaluations.” A quick update for the record.
Language Models (Mostly) Know What They Know
Anthropic posted “Language Models (Mostly) Know What They Know.” A quick update for the record.
Measuring Faithfulness in Chain-of-Thought Reasoning
Anthropic posted “Measuring Faithfulness in Chain-of-Thought Reasoning.” A quick update for the record.
Measuring Progress on Scalable Oversight for Large Language Models
Anthropic posted “Measuring Progress on Scalable Oversight for Large Language Models.” A quick update for the record.
Question Decomposition Improves the Faithfulness of Model-Generated Reasoning
Anthropic posted “Question Decomposition Improves the Faithfulness of Model-Generated Reasoning.” A quick update for the record.
Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned
Anthropic posted “Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned.” A quick update for the record.
Studying Large Language Model Generalization with Influence Functions
Anthropic posted “Studying Large Language Model Generalization with Influence Functions.” A quick update for the record.
The Capacity for Moral Self-Correction in Large Language Models
Anthropic posted “The Capacity for Moral Self-Correction in Large Language Models.” A quick update for the record.
Towards Measuring the Representation of Subjective Global Opinions in Language Models
Anthropic posted “Towards Measuring the Representation of Subjective Global Opinions in Language Models.” A quick update for the record.
Towards Monosemanticity: Decomposing Language Models With Dictionary Learning
Anthropic posted “Towards Monosemanticity: Decomposing Language Models With Dictionary Learning.” A quick update for the record.
Towards Understanding Sycophancy in Language Models
Anthropic posted “Towards Understanding Sycophancy in Language Models.” A quick update for the record.
Frontier Model Security
Anthropic posted “Frontier Model Security.” A quick update for the record.