AI Today: GPT-6.1 Terra and Polar Peaks

State of knowledge as of 2026-10-27. Source material gathered on 2026-10-27.

Today brought releases of educational tools and language models, alongside published data on the use of artificial intelligence in professional work. Google and Stanford released the Polar Peaks game, OpenAI unveiled the GPT-6.1 Terra model and published a report on users' professional habits, CodeRabbit integrated the GPT-6.1 Sol solution with security processes, and students presented practical ways of using the Gemini assistant in their studies.

Highlights Today (TL;DR)

• Polar Peaks: An interactive educational game for students aged 11-14 introducing the basics of neuroscience and AI operations.
• GPT-6.1 Terra: A new OpenAI model featuring improved visual performance and API costs 50% lower compared to its predecessor.
• ChatGPT Work Data: OpenAI released a compilation concerning the model's use across 165 professional categories in nearly 100 countries.
• CodeRabbit: Deployment of the GPT-6.1 Sol model for code dependency analysis and automated vulnerability detection.
• Gemini for Students: A group of Google ambassadors described using the assistant for study organization, exam preparation, and language learning.

Google and Stanford Launch Polar Peaks

Google Research, along with the Stanford Accelerator for Learning, released an interactive educational game called Polar Peaks on October 26. This is the third installment of the AI Quests series, aimed at students aged 11 to 14, designed to introduce the principles of artificial intelligence and the basics of neuroscience.

During gameplay, participants take on the role of researchers mapping the human brain to better understand memory loss in a game character. Students perform tasks related to data cleaning and model testing, designed to show how artificial intelligence supports scientific research. The game mechanics are based on connectomics, the science of mapping neural networks, including a recent achievement: mapping the full fruit fly brain. The program emphasizes the necessity of human oversight over technology and the development of critical thinking.

To date, students have participated in over 400,000 AI Quests sessions, and the materials have been localized in 10 languages. Alongside the game, training modules for teachers were released as part of the Google AI Educator Series, helping integrate this content into the curriculum. Previous installments of the series focused on flood forecasting and eye disease detection. The project won a 2026 Webby Award in the AI education and training category and an ETIH Innovation Award.

For AI builders: Teachers and educators planning to implement the game can check the availability of training modules within the Google educational series. The source does not specify whether the tool complies with GDPR requirements regarding the processing of student data or whether the game features full Polish language interface support.

Release of the GPT-6.1 Terra Model

OpenAI unveiled the GPT-6.1 Terra model on October 26, designed to increase performance in visual tasks and computer-use operations. This model is a progression of the GPT-5.6 Terra version, offering higher processing speeds and a more natural writing style.

According to shared data, GPT-6.1 Terra features API pricing 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. In Agents’ Last Exam tests, at a medium reasoning level, the new model achieved scores 15 percentage points higher than its predecessor at about one-third of the cost. Compared to GPT-6 Astra, the model exhibits comparable benchmark results at one-tenth of the price per token. In terms of speed, GPT-6.1 Terra executes tasks over three times faster than GPT-6 Astra and GPT-6.1 Sol.

In MMMU-Pro visual reasoning tests, the model outperforms GPT-5.6 Terra at every reasoning setting. In programming, on the Terminal-Bench 4.0 test at the maximum reasoning level, GPT-6.1 Terra solves 50.3% of tasks, more than doubling the 21.5% achieved by GPT-5.6 Terra. The manufacturer also points to error reduction: in internal tests, the error rate at the low reasoning level dropped from 53.6% to 16.7%. The model meets the Critical safety threshold under the Preparedness Framework in cybersecurity.

For AI builders: Teams automating financial document processing or desktop application support can test the model's performance in high-volume data entry scenarios. The source does not provide information on the model's availability in specific global regions, Polish language support, or exact token limits under the ChatGPT Work subscription. Deployment decisions should be preceded by verifying detailed licensing terms and interface availability in technical documentation.

OpenAI Shares Data on ChatGPT Workplace Usage

The OpenAI Economic Research team published data on October 26 regarding how the ChatGPT model is used in the professional environment. The dataset covers 165 professional activity categories across nearly 100 countries, allowing for an analysis of AI's role in the economy from the first quarter of 2025 to the third quarter of 2026.

The data is based on the O*NET standard, which groups job tasks into categories called Intermediate Work Activities (IWA). This enables researchers to compare AI usage patterns with other economic indicators and activities performed across various professions. The material includes preparing instructional materials, creating marketing content, programming, and business data analysis.

The shared analyses show that countries tend to specialize in different types of work assisted by ChatGPT. In the third quarter of 2026, the ten most frequently performed tasks accounted for an average of 51.1% of all classified professional prompts, a drop in concentration from the 56.0% recorded in the first quarter of 2025. The dataset comes from private user plans and does not include messages sent by Enterprise users or the use of the Codex model. For privacy protection, the data has been anonymized, and automated systems classify messages after removing identifying data.

For AI builders: Data scientists and market analysts can use these datasets to benchmark specific professional activities against AI usage trends across world regions. Before attempting to integrate proprietary datasets with this data, verify whether the IWA classification methodology matches local labor market specifics and whether the shared indicators account for the full spectrum of business activity, as the source notes the exclusion of tasks like machine operation or medical record keeping.

CodeRabbit Integrates GPT-6.1 Sol into Code Security Processes

CodeRabbit has deployed the GPT-6.1 Sol model as the core of its security agent, helping engineering and application security teams analyze code and detect vulnerabilities. The solution, announced on October 26, aims to extend security controls beyond single code changes, enabling verification of dependencies and component interactions in complex repositories.

The system uses the GPT-6.1 Sol model to combine information about code behavior across various files and libraries. The tool operates independently of the agent that generated the given code, complementing standard scanners. Under the Deep Scan feature, the system analyzes links within repositories, including dependencies in external libraries. CodeRabbit Secure additionally supports threat verification by attempting to reproduce them in an isolated sandbox environment without affecting production systems. When vulnerabilities are detected, the system can prepare a fix proposal as a pull request for human review. According to company data, the solution is currently used to flag security issues in thousands of organizations.

For AI builders: Teams maintaining software infrastructure can test the model's integration into pull request workflows to automate vulnerability detection. The announcement does not specify subscription costs, operational model limits, or whether the system supports security analysis in languages other than those used in the company's existing projects. Deployment requires verifying technology availability in specific regions and data retention policies in the isolated test environment.

Student Use of the Gemini Assistant

According to information published on October 26, a group of 15 Google university ambassadors shared how they use the Gemini assistant in their educational processes. Program participants pointed to the tool's use in study organization, exam preparation, and everyday student duties.

Users leverage the "study notebooks" feature to process course materials into personalized lessons, enabling faster exam review. It is possible to upload notes or lectures and generate comparisons for different viewpoints. The Gemini Live feature enables voice conversations, which students use for practicing presentations, job interviews, or foreign languages. Thanks to camera support, the tool can also guide software usage via visual screen analysis.

Under Guided Learning, instead of providing ready answers, the assistant asks open-ended questions designed to lead the student to independently find errors in a problem's solution. Additionally, Gemini generates knowledge-check quizzes to help prepare for exams. Students also use Google Drive integration, sharing folders with materials from specific subjects to create cohesive course guides.

For AI builders: Educational system creators can examine the mechanism of guiding a user through problem-solving rather than providing direct answers. The press release does not specify upload file format limitations, data privacy rules for shared folders, or whether Guided Learning is available in languages other than English.

Frequently Asked Questions (FAQ)

What are the main performance advantages of the GPT-6.1 Terra model?

GPT-6.1 Terra offers operating speeds over three times higher than the GPT-6 Astra and GPT-6.1 Sol models. Furthermore, the model achieved significant improvements in programming tasks, solving 50.3% of tasks on the Terminal-Bench 4.0 test and showing a clear drop in error rates at low reasoning levels.

How does the Gemini assistant support students in independent study?

Under the guided learning feature, Gemini promotes independent problem-solving by asking open-ended questions instead of providing direct answers. Additionally, students can use study notebooks to create personalized lessons from course materials and generate quizzes to test knowledge before exams.

What does security analysis in the CodeRabbit Secure system involve?

The system uses the GPT-6.1 Sol model to analyze links between files and libraries in repositories. The Deep Scan feature checks component dependencies, and when vulnerabilities are detected, the system can attempt to reproduce threats in a sandbox environment and prepare a code fix proposal as a pull request.

Sources

• Our newest AI Quest: Mapping the human brain — 2026-10-26
• Introducing GPT-6.1 Terra — 2026-10-26
• ChatGPT’s expanding world of work — 2026-10-26
• GPT-6.1 Sol helps CodeRabbit strengthen security across complex codebases — 2026-10-26
• 15 college students share their favorite Gemini features — 2026-10-26

--- Text prepared automatically by the aipoc-lab (AI) laboratory based on publicly available sources listed above. Original summaries, not copies. Pre-human-editing version.