Research Brief

In the first week of February 2026, over $1 trillion in market capitalization was erased from software stocks, triggered by Anthropic's Claude Cowork release on January 12, 2026, a desktop AI agent that lets non-technical knowledge workers offload complex, enterprise-grade tasks to AI. AI poses a structural threat to seat-based models in some use cases and may directly compete with existing products in others. Some vendors are already reporting slower growth in seat count as customer companies become more efficient.

OpenAI released the Codex app for macOS on February 2-3, 2026 (https://openai.com/index/introducing-the-codex-app/). More than a million developers have used Codex in the past month, and usage has nearly doubled since the launch of GPT-5.2-Codex in mid-December, building on more than 20x usage growth since August 2025. The app enables parallel agent workflows with built-in Git worktree support, automations on schedules, and a Skills system for reusable capabilities. GPT-5.3-Codex, released February 5, 2026, advances both the frontier coding performance of GPT-5.2-Codex and the reasoning and professional knowledge capabilities of GPT-5.2, together in one model, which is also 25% faster. GitHub repo: https://github.com/openai/codex.

Anthropic released Claude Opus 4.6 on February 5, 2026. Pricing for Opus 4.6 starts at $5 per million input tokens and $25 per million output tokens. Claude found more than 500 previously unknown zero-day vulnerabilities in open-source code using just its 'out-of-the-box' capabilities. The model features a 1M token context window and supports adaptive thinking with effort parameters.

Alibaba released Qwen3-Coder-Next on February 3, 2026, under an Apache 2.0 license enabling commercial usage, an open-source 80B-parameter coding model that activates just 3B parameters per query, scoring 70.6% on SWE-Bench. GitHub: https://github.com/QwenLM/Qwen3-Coder. Built on top of Qwen3-Next-80B-A3B-Base, which adopts a novel architecture with hybrid attention and MoE, Qwen3-Coder-Next has been agentically trained at scale on large-scale executable task synthesis, environment interaction, and reinforcement learning. Deployment via SGLang (v0.5.8+) and vLLM (v0.15.0+) with tool-calling support.

On February 11th, 2026, Z.ai officially released GLM-5, its new frontier large language model. Live on OpenRouter as of February 11, 2026, it is priced at approximately $0.80–$1.00 per million input tokens and $2.56–$3.20 per million output tokens. GLM-5 uses a Mixture of Experts architecture with DeepSeek Sparse Attention to reduce deployment costs while maintaining long-context capacity. Total Parameters: 744B (40B active) Pre-training Tokens: 28.5T. GitHub: https://github.com/zai-org/GLM-5. Scores 92.7% on AIME 2026 I and 86.0% on GPQA-Diamond. Achieves 77.8% on SWE-bench Verified and 73.3% on SWE-bench Multilingual.

On February 12th, 2026, barely a month after its Hong Kong IPO, Shanghai-based MiniMax dropped M2.5. The headline numbers: 80.2% SWE-Bench Verified, 51.3% Multi-SWE-Bench (first place), 76.3% BrowseComp. M2.5-Lightning has a steady throughput of 100 tokens per second, which is two times faster than other frontier models, and costs $0.3 per million input tokens and $2.4 per million output tokens. M2.5, which has a throughput of 50 tokens per second, costs half that. MiniMax-M2.5 is a Mixture of Experts (MoE) model with 230 billion total parameters, but only 10 billion active parameters are used during inference. MiniMax-M2.5 is released under the MIT license.

GitHub Agent HQ was announced at GitHub Universe in October 2025 and expanded February 4, 2026. Copilot Pro+ and Copilot Enterprise users can now run multiple coding agents directly inside GitHub, GitHub Mobile, and Visual Studio Code. That means you can use agents like GitHub Copilot, Claude by Anthropic, and OpenAI Codex (both in public preview) today. The new architecture transforms GitHub into a unified control plane for managing multiple AI coding agents from competitors including Anthropic, OpenAI, Google, Cognition and xAI. Microsoft Agent Framework: https://github.com/microsoft/agent-framework.

On February 6, 2026, Waymo introduced the Waymo World Model, a frontier generative model that sets a new bar for large-scale, hyper-realistic autonomous driving simulation. The system is built on top of Genie 3, Google DeepMind's general-purpose world model, and adapts it to produce photorealistic, controllable, multi-sensor driving scenes at scale. The model generates both camera and lidar data simultaneously, enabling simulation of rare scenarios like tornadoes, floods, and unusual obstacles that autonomous vehicles seldom encounter in real-world testing.