GPT-6.1 Terra, Gemini 4 Carbon, Mistral Large 4: AI and Developer News for Late October 2026

October 2026 ended with a run of big AI announcements. In a single week, OpenAI shipped a cheaper model built for computer use, Google introduced a new frontier model focused on cybersecurity, Mistral released the full weights of its largest model, and the Gemini API made computer use generally available. Outside AI, Node.js 26 became the new Long Term Support line and GitHub gave Dependabot a smarter way to tell you which security alerts actually matter.

We already covered GitHub Universe and the first Node.js 27 alpha in our October developer news roundup, so this post focuses on the other stories from the same week. Each item links to the official announcement so you can check the details yourself.

OpenAI Releases GPT-6.1 Terra

On October 26, OpenAI introduced GPT-6.1 Terra, a fast, lower-cost model aimed at computer use, visual reasoning, and everyday writing. OpenAI pitches it for the kind of repetitive work that fills a workday, such as reading invoices and entering them into accounting software, filling out forms, or interpreting technical drawings.

The headline for developers is price. OpenAI says Terra's standard API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. Standard pricing is listed at $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens, and the model ID is gpt-6.1-terra.

On coding, OpenAI reports that at maximum reasoning effort Terra solves 50.3% of tasks on Terminal-Bench 4.0, more than double GPT-5.6 Terra's published 21.5%. OpenAI is also clear that for the hardest coding work, GPT-6.1 Sol and GPT-6 Astra "remain the better choice." The company also says it cut the error rate on its internal benchmark of error-inducing prompts from 53.6% to 16.7% at low reasoning effort.

Why it matters: if you are building automations that click through web apps or read scanned documents, a cheaper model that still handles screens and images well can make a project affordable that was not before. As always, test it on your own tasks instead of relying only on vendor benchmarks.

Google Announces Gemini 4 Carbon

Two days later, on October 28, Google announced Gemini 4 Carbon, which it calls its strongest model, with "state-of-the-art cybersecurity and long horizon capabilities." It follows Gemini 4 Argon, which reached general availability in the Gemini API on October 13.

Google's announcement leans heavily on security. It reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. Google also says that internally Carbon has run "for days" on long engineering tasks such as tracking down and patching subtle bugs in large codebases.

The important detail is that you cannot use it yet. Google is rolling Carbon out first to a subset of cyber defenders and then to trusted partners through its Fairwind Program, while it continues to work on guardrails. When it does reach developers, Google says it will start at an introductory price of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period.

Why it matters: the major labs are now openly competing on security work: finding, validating, and fixing vulnerabilities. For students and junior developers, it is a good reminder that secure coding skills are becoming more valuable, not less, as these tools spread.

Mistral Large 4 Is Now Open Weight

On October 29, Mistral released the weights of Mistral Large 4, nicknamed "Le Chonk." The model was first released in public preview earlier in October, and the open-weight version includes further training that Mistral says brought major gains in coding and defensive cybersecurity.

According to the model card on Hugging Face, it is a 1-trillion-parameter, natively multimodal Mixture-of-Experts model with 49 billion active parameters per token (52 billion counting embeddings and output layers). It takes text and images as input. Mistral reports that it improved from its preview score to 41% on Terminal-Bench 4.0, and that on the Artificial Analysis Coding Agent Index it is effectively tied with GLM-5.3 while using 47% fewer tokens per task on average.

You can use it through the Mistral API (mistral-large-4) at $0.70 per million input tokens and $2.10 per million output tokens, or download the weights and run them yourself with vLLM or SGLang. The Hugging Face launch post also mentions day-zero support in transformers and llama.cpp, plus GGUF builds.

Check the license before you build a business on it. The Mistral Ecosystem Open-Weight License is a modified MIT license: you can use, modify, fine-tune, and commercialize the model, but companies that run a Model-as-a-Service offering need a separate commercial license from Mistral.

Why it matters: a trillion-parameter model is far beyond a laptop or a typical school lab, but open weights still matter. Companies and governments that cannot send data to an outside API now have another strong option, and the open ecosystem tends to produce smaller, cheaper variants over time.

Gemini API: Computer Use Goes GA and a New CLI

The Gemini API release notes had two items worth knowing about:

• October 30: Computer Use is generally available across both the Interactions and GenerateContent APIs. This is the tool that lets a model operate a browser or app interface, and GA status means it is now considered ready for production use.
• October 27: Gemini API CLI (gemini-api) in beta. Google describes it as a stateless, zero-dependency command-line client for AI coding agents, scripts, and CI pipelines. It can generate text and media, analyze files, and run Gemini managed agents.

One more change from October 26 is easy to miss: new paid-tier projects using the Generate Content API now store requests by default (store=true). Existing projects keep their current setting. If you work with sensitive data, read the note in the release log and opt out with store=false or in the logging settings.

Node.js 26 Becomes the New LTS

On October 28, the Node.js project released Node.js 26.13.0, which moves the 26.x line into Long Term Support with the codename "Lithium." It will receive updates until the end of April 2029. The release itself contains no new features compared with 26.12.1. It only marks the line as LTS.

The release notes include one practical warning if you are upgrading from Node.js 24, 22, or older: since Node.js 25, the official Linux binaries are linked against libatomic, which you may need to install separately depending on your distribution. If your production servers or Docker images still run an older LTS, now is a good time to plan the upgrade and test it.

Dependabot Can Now Tell You Which Alerts Are Reachable

Also on October 28, GitHub introduced reachability analysis for Dependabot in public preview. Anyone who has opened a repository with dozens of Dependabot alerts knows the problem: a vulnerable package in your dependency tree does not always mean your code uses the vulnerable part.

Reachability analysis uses CodeQL to build a call graph of your code and its dependencies, then gives each alert a verdict: Reachable, Potentially reachable, Unreachable, or Reachability unavailable. The verdicts never close or dismiss an alert. They only help you decide what to fix first.

During the preview it is free, but it is limited to GitHub Enterprise Cloud and to JavaScript and TypeScript repositories with an npm, Yarn, or pnpm lockfile. It is off by default and is turned on from the Advanced Security settings. If you are not on Enterprise Cloud, you can still audit your dependencies yourself with free command-line tools. Our guide on finding vulnerable dependencies with npm audit, pip-audit and composer audit shows you how.

What This Means for You

• AI pricing keeps falling. GPT-6.1 Terra and Mistral Large 4 both compete hard on price. If you shelved an AI feature because it cost too much, run the numbers again.
• Security is the new battleground. Gemini 4 Carbon, Mistral Large 4, and Dependabot reachability all focus on finding and fixing vulnerabilities. Learning secure coding basics is a smart career move.
• Agents are leaving the chat window. Computer use is now generally available in the Gemini API, and it is a main selling point for Terra. Expect more tools that click, type, and fill in forms for you, and design your apps so they can be tested safely that way.
• Plan your Node.js upgrade. Node.js 26 is the LTS line to target for new production work.

Conclusion

The last week of October showed where the industry is heading: cheaper models that can operate computers, frontier models built for security work, open weights at a scale few expected, and smarter tools for sorting through security alerts. You do not need to adopt everything at once. Pick the one item that affects your current project, read the linked announcement, and try it on a small piece of real work first. That is still the best way to separate real progress from hype.