This week in AI updates: GPT-5.3-Codex-Spark, GitHub Agetic Workflows, and more (February 13, 2026)

OpenAI releases research preview GPT-5.3-Codex-Spark for ChatGPT Pro users

GPT-5.3-Codex-Spark is a lightweight version of the company’s coding model, GPT-5.3-Codex, optimized to run on ultra-low latency hardware and can deliver over 1,000 tokens per second.

It is the first result of the company’s recently announced partnership with Cerebras to add 750MW of ultra low-latency AI computing. OpenAI says the reason it’s being released as a research preview is to give developers a chance to start benefiting from this partnership while the company continues to roll out Cerabras in its data centers.

“Codex-Spark is our first model designed specifically to work with Codex in real time—to make targeted edits, refactor the mind, or adjust interfaces and see results quickly. With Codex-Spark, Codex now supports both long-term, ambitious tasks and getting work done in the moment. We hope to learn from how developers use it and continue to expand the post as AI writes.

GitHub is launching a technical preview of Agentic’s workflow

GitHub Agentic Workflows allow developers to define the desired outcome in transparent Markdown, add it as an automated workflow to their repository, and have that implemented as a code agent in GitHub Actions.

Agentic Workflows work like standard GitHub Actions workflows with additional monitoring lines for sandboxing, permissions, control, and review. Additionally, they support a variety of agent coding engines, including Copilot CLI, Claude Code, or OpenAI Codex.

“Using GitHub Agentic Workflows enables entirely new phases of repository automation and software engineering, in a way that fits naturally with how development teams already work on GitHub,” GitHub wrote in a post.

Google Adds Automatic Updates to Conductor in Gemini CLI

Conductor is a Gemini CLI extension that helps bring more development context to the terminal, and the new Auto Update feature generates a post-launch report after the agent has completed its coding tasks. Findings are categorized by severity (low, medium, and high) so developers can prioritize where to iterate first.

“This level of detail ensures that a “functional” upgrade does not mean an ‘unsupervised’ upgrade.” Instead, it creates a workflow where AI provides the work and the developer provides high-level architecture guidance, supported by automated validation,” Google wrote in a blog post.

Additionally, Google announced that Gemini CLI extensions will now be able to define the settings that the user will be prompted to provide when installing the extension. By providing things like API keys, base URLs, and project identifiers up front, users can hopefully have fewer configuration errors to troubleshoot when working with the Gemini CLI, the company explained.

Google is developing Gemini 3 Deep Think mode

According to Google, the advanced model includes advances in all mathematical and programming thinking methods, as well as certain scientific fields such as chemistry and physics. Available to Google AI Ultra subscribers on the Gemini app, and to select researchers, developers, and businesses on the Gemini API for the first time.

“We updated Gemini 3 Deep Think in collaboration with scientists and researchers to address difficult research challenges – where problems often do not have clear rules of thumb or a single appropriate solution and data are often dirty or incomplete. By combining deep scientific knowledge with everyday engineering services, Deep Think goes beyond abstract theory to drive applications,” Google wrote in a post.

Testing GitHub Copilot for .NET is now generally available

Now available in Visual Studio 2026 v18.3, the test feature allows developers to generate unit tests within their IDE. The company has added new capabilities to accompany this GA release, including deeper IDE integration, more intuitive prompts, and new ways to request test information.

The team plans to focus next on making it handle more advanced test applications, which will include addressing requirements such as allowing engineers to specify intent, validate assumptions, and review proposed systems before producing tests.

“General availability is an important milestone, but it’s not the end of the journey. We continue to conduct user studies and gather feedback to understand how developers are using GitHub Copilot testing for .NET in real-world situations, especially as applications grow in size and complexity,” Microsoft wrote in a blog post.

Anthropic raises $30 billion in Series G funding

This latest round of financing was led by GIC and Coatue; co-led by DE Shaw Ventures, Dragoneer, Founders Fund, ICONIQ, and MGX; and contributed to 30 other investors. As of this round, Anthropic is now valued at $380 billion post-money.

In its announcement, Anthropic also revealed that its run-rate revenue is now $14 billion, and has grown 10x every year since its initial funding three years ago.

“Whether they are entrepreneurs, startups, or the largest enterprises in the world, the message from our customers is the same: Claude is becoming more and more serious about how businesses operate,” said Krishna Rao, Anthropic’s chief financial officer. “This fundraising reflects the incredible demand we’re seeing from these customers, and we’ll use this investment to continue building the enterprise-class products and models they’ve come to rely on.”