AI News of the Week (6th February, 2026)
TLDR: Key AI Developments This Week
OpenAI launched a native Codex for macOS app and GPT-5.3 Codex model. Anthropic released Claude Opus 4.6 with a 1M context window and "Agent Teams," plus deep integrations with PowerPoint and Excel. GitHub Copilot now supports Claude models. OpenAI announced the Frontier enterprise platform. Leaks suggest Claude 5 "Fennec" is coming soon, and DeepSeek V4 is expected in mid-February.
OpenAI Launches Native "Codex for macOS" App
OpenAI has released a standalone Codex for macOS application, moving beyond the browser and IDE extensions. This native app integrates deeply with the macOS accessibility API to "watch" your local development environment, offering system-wide code suggestions and terminal command auto-corrections. It supports local indexing of your entire file system (with privacy toggles), allowing it to understand project context without manual uploads.
OpenAI Releases GPT-5.3 Codex Model
Alongside the Mac app, OpenAI deployed GPT-5.3 Codex, a specialized version of their frontier model fine-tuned exclusively for software engineering. The model boasts a 92% score on the new SWE-bench "Hard" dataset and introduces "lazy loading" for context—it only retrieves definitions from your repo when specifically needed for the query, reducing token costs by 40% while maintaining perfect recall for large codebases. The model is 25% faster than GPT-5.2 Codex and helped build itself during development.
Anthropic Releases Claude Opus 4.6 with 1M Context Window
Responding to OpenAI's push, Anthropic unexpectedly dropped Claude Opus 4.6. This update focuses heavily on "agentic reliability" and "self-correction." In benchmark tests, Opus 4.6 demonstrated the ability to catch its own logic errors during complex architectural planning 9 times out of 10 without human intervention. The model now supports a 1-million-token context window (beta), eliminating the "context rot" problem that plagued previous models. On GDPval-AA (economically valuable knowledge work tasks), Opus 4.6 outperforms GPT-5.2 by 144 Elo points and scores 80.9% on SWE-bench, leading the industry in coding benchmarks.
Anthropic Launches Claude in PowerPoint (Research Preview)
Alongside Opus 4.6, Anthropic released Claude in PowerPoint as a research preview for Max, Team, and Enterprise plan customers. The integration allows Claude to build, edit, and refine presentations directly within PowerPoint. Claude reads your slide masters, layouts, fonts, and color schemes to stay on-brand. It can convert data from Excel into professional visualizations, create entire decks from descriptions, or make pinpoint edits to existing slides. The tool generates editable native PowerPoint charts and diagrams, not static images.
Anthropic Upgrades Claude in Excel with Opus 4.6
Anthropic significantly upgraded Claude in Excel to use Opus 4.6. The enhanced version can now handle native Excel operations including pivot table editing, conditional formatting, and complex multi-tab workbook analysis. Claude can ingest unstructured data and infer the right structure without guidance, plan before acting on complex tasks, and handle multi-step changes in one pass while preserving formula dependencies. It reads complex calculations and provides cell-level citations when explaining workbook logic.
Anthropic Introduces Agent Teams (Research Preview)
Claude Opus 4.6 introduces a research preview feature called "Agent Teams" for multi-agent collaboration. This experimental feature allows developers to spawn multiple sub-agents that work in parallel on complex development tasks. These sub-agents can verify and peer-review each other's work, significantly boosting code quality. The feature is token-intensive and requires setting CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. Additionally, Claude now automatically records and recalls memories as it works, maintaining context across long sessions.
GitHub Copilot in VS Code v1.109 - January 2026 Release
GitHub released significant improvements to Copilot in Visual Studio Code. The January 2026 release introduces agent support for Claude by Anthropic (public preview), allowing developers to delegate tasks directly to Anthropic's Claude Agent SDK using Claude models from their Copilot subscription. The update includes richer chat interactions with MCP apps, improved streaming responsiveness, higher-quality reasoning results, revamped editor inline chat, Agent Skills with org-wide customization, Copilot Memory for retaining context, faster code search via external indexing, and experimental terminal command sandboxing (macOS/Linux). An integrated browser was also added for testing apps without leaving the editor.
OpenAI Launches Frontier Enterprise AI Platform
OpenAI announced Frontier, an enterprise platform designed to build, deploy, and manage AI agents across systems with shared context, onboarding, feedback, and strong boundaries. The platform acts as a "semantic layer for the enterprise" that all AI coworkers can reference to operate effectively. Frontier is built on open standards (including Model Context Protocol), allowing software teams to plug in and build agents that benefit from shared business context. Early adopters include partners like Abridge, Clay, Ambience, Decagon, Harvey, and Sierra.
Claude 5 "Fennec" Leak Surfaces in Vertex AI Logs
The AI community discovered evidence of Claude 5 in Google Vertex AI deployment logs. A model identifier claude-sonnet-5@20260203 appeared in error logs, suggesting Anthropic is testing Claude Sonnet 5 (codenamed "Fennec") internally. While Anthropic hasn't officially confirmed the model, cross-verification from multiple independent sources indicates it's legitimate. Industry analysts predict Claude 5 (or at least Sonnet 5) will likely launch in February or March 2026, with expected improvements including 50% lower pricing than Sonnet 4.5, multi-agent "Dev Team mode" for collaborative development, and even stronger safety alignment.
DeepSeek V4 Coding Model Expected Mid-February
Chinese AI startup DeepSeek announced plans to launch DeepSeek V4, a coding-optimized model, around mid-February 2026 (likely February 17, coinciding with Lunar New Year). The model introduces a revolutionary "mHC architecture" and supports 1M+ token context windows. Internal testing reportedly shows V4 outperforming Claude 3.5 Sonnet and GPT-4o on coding benchmarks, though independent verification is pending. DeepSeek plans to release V4 as an open-weight model, continuing their tradition of making powerful AI accessible to the broader community. The model is optimized to run on dual consumer GPUs, making state-of-the-art coding AI accessible for local deployment.
Anthropic Announces Data Residency Controls
Anthropic introduced data residency controls for enterprise customers, allowing them to specify where model inference runs using the inference_geo parameter. US-only inference is available at 1.1x pricing for models released after February 1, 2026. This addresses regulatory requirements for companies in highly regulated industries like finance and healthcare. The feature is available across all Claude models and platforms.
