Baseten vs OpenRouter

Goodfire Launches Internal AI Agent Monitors

Goodfire, a startup specializing in AI interpretability, launched on Thursday a new type of AI agent monitor that inspects a model's internal signals rather than reading its output, aiming to detect rogue behaviors more efficiently and at a fraction of the cost. The monitors are available to customers of Baseten, an AI model hosting platform. Baseten's Base Labs had previously announced a safety partnership with Goodfire and Hugging Face. Goodfire's approach uses small probes that scan a model's internal activations at each step, triggering a closer AI review only when flagged. In tests on the Kimi K3 model, Goodfire's monitors caught 94% of malicious hacking sessions and cost about $51 for 1,500 sessions, compared to $233 for a cheaper model and $10,000 for a top-tier one. The company positions the solution for open models, which can be stripped of safeguards, and sees it as critical for inference-time guardrails.

• Goodfire launched internal activation monitors for AI agents on October 8, 2026.
• Monitors are available to Baseten customers.
• In tests on Kimi K3, monitors caught 94% of malicious hacking sessions.

Announcing our partnership with OpenAI

Baseten announces a partnership with OpenAI, as featured on their blog.

Baseten, Hugging Face, Goodfire Partner for AI Safety

Baseten's research arm, Base Labs, announced a partnership with Hugging Face and Goodfire AI to establish a safety standard for open-weight AI models. The collaboration aims to develop and publish methods for training and monitoring these models, making safety an integrated feature rather than an afterthought. This comes amid rising concern over 'abliteration,' a technique that removes safeguards from open models, with Hugging Face listing over 6,000 such models. The companies have not detailed technical specifics, but Goodfire's expertise in AI interpretability is expected to play a key role. This initiative seeks to leverage openness as an advantage for safety, providing transparent controls and encouraging community contributions.

• Baseten launched a safety infrastructure standard via Base Labs.
• Partnership with Hugging Face and Goodfire AI announced.
• Hugging Face hosts over 6,000 abliterated models.

ONLYOFFICE 10.0 Launches with AI Tools, Modular Apps

ONLYOFFICE has released version 10.0 of its open-source office suite, introducing context-aware AI tools embedded directly into its editors. The update adds a dedicated AI tab with a chat interface for drafting, summarizing, translating, and formatting content. Organizations can connect various AI models via OpenRouter or local deployments, ensuring data control. Four new modular apps—AI Agents, Forms, Files, and Rooms—extend ONLYOFFICE Docs, offering automation, document management, and secure collaboration, with AI Agents powered by MCP servers. The release includes significant security improvements, fixing 35 vulnerabilities, plus editor enhancements across documents, spreadsheets, presentations, and PDFs, including metadata removal and reusable signatures. Desktop and mobile apps receive refreshed interfaces, Odoo integration, and mobile AI assistants supporting providers like Grok and DeepSeek, along with AI form autofill and voice input. With over 21 million users, ONLYOFFICE strengthens its position in secure office productivity.

• ONLYOFFICE 10.0 introduces context-aware AI tools with a dedicated AI tab, supporting cloud or local AI models.
• Four new free modular apps (AI Agents, Forms, Files, Rooms) are added under AGPL v3, with AI Agents powered by MCP servers.
• The update fixes 35 security vulnerabilities and includes over 1,000 fixes across editors.

Firebase Global Outage, OpenAI Platform Shift

This issue of The Pulse covers the global outage of Firebase's iOS SDK, which crashed apps using Firebase Analytics for 2-6 hours. Google failed to update its status page or provide a postmortem, criticized given its reputation for incident management. Also, OpenAI is evolving into a platform where users can allocate ChatGPT spend across open models and AI offerings from 16 partners, not just OpenAI's own models. Data from Vercel's AI gateway shows 60% of model spend goes to open-weight models, with OpenRouter confirming a shift towards open models. Additionally, a new take suggests CTOs and VPEs are leaving due to AI tools enabling solo or small-team development, rather than 'founder mode'.

• Firebase iOS SDK crashed, affecting apps for 2-6 hours.
• Google did not update status page or publish a postmortem.
• OpenAI now allows ChatGPT spend on open models from 16 partners.

Claude Opus 5.5 Dominates Explainer Video Creation

Anthropic's Claude Opus 5.5 has become a sensation in the AI community for its ability to generate high-quality motion graphics and explainer videos entirely from code. Users report creating impressive 15-second to 90-second animated videos with single prompts, leading to viral posts on X and Reddit. The model tops OpenRouter in share of spend and tokens among Anthropic models. Beyond video generation, Opus 5.5 leads benchmarks like SimpleBench at 88.4% and is praised as Anthropic's best vision model. The community notes its cost-efficiency, being 60% lower cost than Fable 5.1. The newsletter also covers other AI developments including GPT-6 variants, Gemini 3.8 Flash, and Xiaomi's MiMo-V2.6-Pro, alongside new decision models like TypeSafe's Jev and infrastructure updates from LangChain and Perplexity.

• Opus 5.5 is the #1 model on OpenRouter in share of spend and tokens among Anthropic models.
• Opus 5.5 leads SimpleBench at 88.4% and is ranked Anthropic's best vision model.
• Users create viral explainer videos with Opus 5.5 using single prompts, with videos exceeding 2 million views.