OpenAI releases GPT-6.1 Terra, a cheaper mid-tier model for computer-use agents

·10 views

OpenAI launched GPT-6.1 Terra on October 26, 2026, a mid-tier model designed to operate computers by reading screens, clicking, typing, and switching between apps to complete multi-step tasks. Priced at $1 per million input tokens and $5 per million output tokens, it costs half as much as its predecessor, GPT-5.6 Terra. The model is available to paid ChatGPT Work and Codex users, on Amazon Bedrock, and in GitHub Copilot, while free and Go plans continue to use GPT-5.6 Terra. OpenAI's performance claims, including gains on professional computer tasks and terminal benchmarks, are self-reported and have not yet been independently verified. The company positions Terra for high-volume routine automation, with its more powerful Sol and Astra models reserved for the hardest coding and reasoning work.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Hugging Face Benchmarks Update: 29 New Entries and Shifts at Top Spots

Hugging Face's official benchmark tracker, which pulls data from its public API, reports 57 official benchmarks and 1,169 leaderboard entries as of November 1, 2026 (KST). New #1 positions were recorded on several boards, including ParseBench, SWE-bench_Pro, BRIGHT, decision-index-leaderboard, arguana, and terminal-bench-science. Six new official benchmarks were added, and 29 new leaderboard entries were logged across all boards. FINAL-Bench holds the most #1 spots with 14, followed by deepseek-ai with 5 and zai-org with 5.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

n8n 2.45.7 stable patch requires execution:read for workflow-scoped execution data routes

n8n released patch versions 2.46.3 on its beta line, 2.45.7 on stable, and 1.123.91 on the legacy 1.x series on 30 October. Only the 2.45.7 release notes list a security hardening fix, which makes workflow-scoped routes enforce the execution:read scope before returning execution data. Previously, users whose roles covered workflow scopes but not execution scopes could receive that data through these routes. A related change aligns the AI Assistant so it also requires execution:read to read executions or add recent runs to its context. No GitHub security advisory for this change appears as of publication, and operators are advised to upgrade and review custom roles, since integrations relying on the old behavior will need the execution:read scope added explicitly.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Locentra: Open-Source Lab for Simulating Indoor Positioning Algorithms

Locentra is an open-source, interactive platform for researching how wireless systems estimate the positions of people and objects indoors, where GPS is unreliable. It offers a 2D and 3D virtual lab for rooms and anchors, and simulates real-world issues such as noise, packet dropouts, non-line-of-sight bias, and outliers. Users can compare algorithms including non-linear least squares, Huber robust estimation, and Kalman filtering, using metrics like RMSE, MAE, and P95 error. The project is built with React, Three.js, Python, FastAPI, and NumPy/SciPy, and is hosted on Vercel and Render. Its developers say the aim is to make indoor localisation research more transparent, reproducible, and accessible without expensive hardware.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Linux Kernel TCP Flaw CVE-2026-98432 Rated CVSS 9.8, Patches Lagging on Some Distros

A use-after-free vulnerability tracked as CVE-2026-98432 and rated CVSS 9.8 has been confirmed in the Linux kernel's IPv4 TCP connection handling. It stems from a race condition in which IP options are transferred by pointer during handshake completion, allowing memory freed on one CPU to be read by another. Affected kernels date back to version 4.4, and a two-line fix by Google's Eric Dumazet, reported by Anthropic's Xinyang Ge, was merged on October 5. Anthropic said an unreleased internal model built a working exploit granting unauthenticated root access on an MPTCP listener within 53 minutes. As of October 30, Debian and Oracle Linux had fixes while Ubuntu and RHEL patches were still unavailable, so blocking TCP packets with IP options using nftables was recommended as an interim measure.

0 comments Read more at DEV Community