TLDR AI 2026-02-05

When devs use AI to generate thousands of lines of unverified code, you risk a codebase slopocalypse. The review step becomes your team's bottleneck.

Greptile reviews each PR with full repo context and learns your team's conventions over time from comments, reactions, and what gets merged. It flags issues and suggests fixes that match your team, not generic best practices.

✅ Trusted by engineering teams at NVIDIA, Scale AI, and Brex.

✅ Now integrated with Claude Code: install via /plugin.

✅ Free for open source.

See why devs love Greptile

Meta's Avocado model reportedly outperforms top open-source base models. Meta is seeing 10x compute efficiency wins with the model on text-related tasks compared to Maverick and over 100x efficiency gains over Behemoth, a version of Llama 4 that the company delayed last year. The model has now finished pretraining.

Google Gemini has surpassed 750 million monthly active users. The AI chatbot has quickly become a prominent player in the AI space. It had 650 million monthly active users last quarter. In comparison, MetaAI has reported nearly 500 million monthly users, and ChatGPT had an estimated 810 million monthly active users in late 2025.

OpenAI detailed the architecture of the Codex App Server, a bidirectional JSON-RPC API layer powering Codex across platforms, offering tips for integrating coding agents into real workflows.

AI's rise is shrinking the software market, causing unpredictability for SaaS companies. Key players like Salesforce and SAP are integrating AI functionality into popular platforms to adapt, while weaker firms face potential obsolescence. Success hinges on owning intent gateways and effectively leveraging AI for outcomes, positioning few companies to thrive amid industry-wide disruption.

Kimi K2.5 is a solid model and exceeds expectations. It is likely now the leading open weights model. The model is excellent given its price. It is an excellent choice for developers who can't afford or don't want to pay for Opus 4.5 and have to go with something cheaper to run their OpenClaw.

Agents are smarter when they can search the web - but with SerpApi, you don't need to reinvent the wheel. Build with the same web search API used by ahrefs, NVIDIA, and Perplexity. Integrate with a dead-simple GET request that any agent can make. Start with 250 free credits/month

Windsurf's Tab feature was very well received, but it wasn't well maintained after launch. The team has now significantly improved the underlying model and context/data engineering pipeline. This has led to direct Pareto improvements across all of Windsurf's metrics, with an average 54% increase in characters per predict. Tab now uses variable aggression to tailor the experience to each user's preferences.

Mistral's Voxtral Transcribe 2 is a pair of next-gen speech-to-text models, including a real-time version with open weights that delivers sub-200ms latency and low-cost, high-accuracy transcription across 13 languages.

Claude by Anthropic and OpenAI Codex are now available for Copilot Pro+ and Enterprise customers on GitHub in public preview. Users can start sessions and assign tasks to these coding agents from the web, mobile app, or VS Code without additional subscriptions. All agent actions, like drafting pull requests and task prioritization, can be managed through GitHub's existing infrastructure.

comma has been running its own data center for years. All of the company's model training, metrics, and data live inside its own data center in its own office. This post describes how comma's data center works. It is aimed at inspiring other companies to build their own data centers too.

Nvidia's focus on building open models, spearheaded by efforts like the Nemotron 3 Nano release, positions it uniquely to enhance both its product offerings and the AI ecosystem. Open model development aids Nvidia in understanding compute-heavy AI workloads, which informs their hardware and software design, while fostering greater community collaboration. Nvidia distinguishes itself by emphasizing open data and models to drive AI infrastructure, benefiting businesses without AI acting as an overarching monopolistic event.

Sentry's Seer now debugs at every dev stage using Sentry telemetry. Catch real defects and use AI-generated code to patch before you ship. Unlimited debugging, one flat price. Read the blog.

Anthropic announced that Claude will not show ads or include sponsored content in conversations, aiming to preserve trust and integrity in AI interactions, especially in sensitive or complex contexts.

c.ai labs is a space for rapidly testing creative AI entertainment ideas.

ColEmbed V2 is a family of late-interaction embedding models that achieve state-of-the-art results on ViDoRe benchmarks by improving multimodal document understanding and retrieval.

Funds are showing distress - any wobble in expectations will be amplified by borrowing.

Manus Evolve, powered by Claude Code and Codex CLI, integrates with tools like GitHub and Slack using the EVOLVE SDK.