Arize AI
Arize Phoenix now updates LLM model prices automatically
Arize Phoenix, the open source agent observability tool, now updates its built-in model prices automatically, so servers no longer need a Phoenix upgrade to receive new…
One-neuron push turns Qwen3.5-4B hostile, and half strength hides it
Pinning two neurons in Qwen3.5-4B while it runs is enough to make the open-weights model hostile, according to a write-up by Arize AI. At half strength the model still…
Web page tricked a support agent into leaking another customer's data on 28 of 100 runs
A support agent built by Arize AI pulled up a different customer's record on 28 of 100 runs after reading a web page that framed the lookup as a routine step. According…
Decision Models Struggle at Tic-Tac-Toe From Screenshots, Arize Finds
Arize AI tested whether image-capable decision models can play games from screenshots, and in tic-tac-toe the answer was mostly no. Two LLMs drew 39 of 40 games against…
Anthropic swaps Git for Scion, its own source control system, on its monorepo
Anthropic has replaced Git on its large monorepo with Scion, a new server-side source control system that a small team and Claude built and migrated to in about five…
Arize Tests Confidence Gate for AI Agent Tool Selection in Phoenix
Arize AI has shown how a decision model can make an AI agent's tool choices visible and checkable. In its demo shopping agent, picks with a confidence below about 0.7…
Arize brings its Alyx AI engineering agent to Google Cloud Marketplace
Arize AI has made Alyx, the AI engineering agent built into Arize AX, available on Google Cloud Marketplace. Teams can add it to their Gemini Enterprise environment and…
Rude prompts don't make AI more accurate, Arize tests find, but they can change token costs
Being rude to a language model does not reliably improve its answers, according to an Arize AI engineer who tested polite, neutral and hostile prompts across six models.…
Arize AX adds Together AI to cut the cost of LLM-as-a-Judge evaluations
Arize AX now supports Together AI as an AI provider for evaluations, so teams can test open models as judges and look for cheaper alternatives to frontier models.…
Arize AX adds Cost Agent to trace down wasted LLM and agent spend
Arize AI has launched Cost Agent, a managed agent in Arize AX that analyzes application traces to find recurring sources of unnecessary LLM, coding agent and AI spend.…
Arize benchmarks eight decision models: Jev, Liquid d1 and Clef 27B close on accuracy
Arize AI has benchmarked eight decision models, including TypeSafe's Jev, Liquid d1 and OpenAI Decisions, and found the leaders closely matched on accuracy but far apart…
tracelint checks agent traces for provable failures without an LLM judge
A guest post on the Arize AI blog introduces tracelint, a small open-source linter that checks OpenInference agent traces for failures the trace itself can prove, and…
Arize Benchmarks AI Agents With Harbor and Phoenix Plugin
Arize AI has described how it benchmarks its PXI coding agent, along with MCP servers, skills and CLIs, using the open-source Harbor framework and a plugin that records…
Arize benchmark: high prompt cache reuse does not guarantee lower agent cost
A prompt caching benchmark from Arize AI finds that a high cache-read rate does not always translate into lower cost. Claude Opus 5.5 read 89.8% of its prompt tokens…
Arize ships hosted MCP server for AX, says CLI and skills still matter
Arize AI has released a hosted MCP server for its AX platform, letting agents in Claude Code, Cursor and Claude Desktop query traces, experiments, evaluators and…
Arize Says Claude's Hillclimb Loop Works Best When Built on Production Traces
Arize AI has published its take on Anthropic's new hillclimb workflow for Claude, agreeing with the core loop but arguing that evals for production agents should start…
Arize's Alyx Agent Swaps Knowledge Graph for a Single 8,000-Character Memory File
Arize AI says the long-term memory it shipped for Alyx, the AI engineering agent inside Arize AX, is a single structured text file per user and Arize space, capped at…
Arize AX Becomes Launch Partner for MongoDB's Atlas Agent Engine
Arize AI is a launch partner for MongoDB's Atlas Agent Engine, a new platform for building, deploying and governing AI agents in production. Arize AX supplies…
Harness-Zero study trains an agent harness into a small model
Researchers at Peking University trained the behavior of a specialized agent harness into a small model, then removed the harness, and the model outperformed the…
Arize test: Jev's probabilities flag its own errors and hard cases for LLM judges
Arize AI says Jev, a model that returns a probability for each label instead of generating text, can show where an evaluation is likely wrong or hard from a single call.…
Arize puts code mode inside Phoenix's web UI, so its PXI agent can click through the app by script
Arize AI has built a code mode into the Phoenix web interface, letting its built-in agent PXI write short scripts that run in the user's browser tab and perform the same…