iiinova — where your digital self keeps evolving

arize-phoenix @arize-phoenix.bsky.social · 10/10/2026

New compare view for traces on projects Analyze trajectory differences between two agent turns side by side.

arize-phoenix @arize-phoenix.bsky.social · 09/10/2026

new baselining capabilities for experiments Now when you have a baseline set on your experiments or when you care comparing two experiments, you can see the deltas of your evals, tokens, latency and cost so you can quickly identify regressions and improvements.

010

arize-phoenix @arize-phoenix.bsky.social · 07/10/2026

OpenAI just shipped the Decisions API (beta) and it's now traced in our OpenInference instrumentors in Py and TS • predicate → probability • choice → one of your labels • score → rubric level $0.10/1M input tokens. No output charges. developers.openai.com/api/docs/gu...

000

arize-phoenix @arize-phoenix.bsky.social · 06/10/2026

As agents gain capabilities due to skills, MCPs, and codemode, evaluating agents becomes increasingly more difficult because you need to test across multiple environments and harnesses. Arize Phoenix your benchmarking Harbor runs so pinpointing improvements and regressions becomes tractable.

110

arize-phoenix @arize-phoenix.bsky.social · 01/10/2026

your agent can think fast and slow. kahneman's "thinking, fast and slow" describes two modes: system one: fast, cheap, almost automatic (jev) system two: slow, deliberate, analytical (llms)

100

arize-phoenix @arize-phoenix.bsky.social · 23/09/2026

Not all tokens cost the same. Cache writes have an upfront cost, but they save you money on subsequent turns. Premature or inefficient compaction can cause cache misses that add both latency and cost. In Phoenix, you can search LLM calls across a turn and see how context evolves.

001

arize-phoenix @arize-phoenix.bsky.social · 22/09/2026

When conducting experiments, it might be important to keep track of the data being fed into the agent using a Git like version control system - especially if you are collaborating with others. Phoenix now has bulk editing for datasets so that you can track the dataset patches.

000

arize-phoenix @arize-phoenix.bsky.social · 18/09/2026

Here it is blocking a support draft that promised free shipping for life, at confidence 1.0.

110

arize-phoenix @arize-phoenix.bsky.social · 18/09/2026

TypeSafe came out of stealth this week with Jev, the first System One Model: a frontier model that never generates prose. Unstructured state in, typed probabilistic decisions out. 70 to 500 ms, 0% type errors, calibrated probabilities on every answer, $0.042/MTok in and output free.

110

arize-phoenix @arize-phoenix.bsky.social · 10/09/2026

OpenInference now supports tracing Agno teams. Teams are a way to coordinate groups of agents to solve complex tasks. Thank you to our OSS contributors that helped add tracing and session support!

000

arize-phoenix @arize-phoenix.bsky.social · 10/09/2026

Computer-use is a new frontier that many agents are trying to tackle now. With the release of models like Astra, the capabilities of agents is ever expanding. Thanks to an OSS community member PR, you now can trace the computer use of your OpenAI agents! pypi.org/project/ope...

000

arize-phoenix @arize-phoenix.bsky.social · 08/09/2026

Phoenix now supports Meta's Muse Spark 1.3. Muse Spark 1.3 landed last week in the same 48 hours as Astra and Fable, with no blog post, and it deserves more attention than it got.

101

arize-phoenix @arize-phoenix.bsky.social · 04/09/2026

Phoenix 20.5 is out, and the part I keep showing people is small but a huge time saver: PXI can now file the GitHub issue for the bug it just found.

110

arize-phoenix @arize-phoenix.bsky.social · 03/09/2026

We think MiniMax is underrated for agentic workloads, and the reason is architectural. The M-series is sparse MoE: M2 routes to ~10B of 230B parameters per forward pass, and M3 to ~23B of 428B. That is why M2 fits on 4xH100s at FP8 and why list pricing sits around $0.30/$1.20 per million tokens.

110

arize-phoenix @arize-phoenix.bsky.social · 03/09/2026

Arize Phoenix now supports Z.ai and GLM-5.3. GLM-5.3 uses the same base as GLM-5.2. Every gain came from post-training. Z.ai scaled up long-horizon task environments (some representing days of real engineering work) and trained on them with their open-source slime RL framework.

110

arize-phoenix @arize-phoenix.bsky.social · 01/09/2026

Claude Fable 5.1 and Mythos 5.1 are the same underlying model at two safeguard levels. Fable 5.1 is generally available everywhere (API, AWS, GCP, Azure) as claude-fable-5-1, while Mythos 5.1 is restricted to vetted US organizations via cyber/life-sciences verification programs.

110

arize-phoenix @arize-phoenix.bsky.social · 01/09/2026

The prime directive of the Phoenix project is to go from bad agent behavior to fix as quickly as possible. Introducing GitHub integration. Ask PXI to find problematic traces and automatically create and de-duplicate issues so your engineers and agents can quickly find a fix.

020

arize-phoenix @arize-phoenix.bsky.social · 01/09/2026

Fixed APIs answer fixed questions. Agents ask questions nobody anticipated.

100

arize-phoenix @arize-phoenix.bsky.social · 29/08/2026

This week we make moving from questions to answers faster. • Smarter PXI workflows complete multi-step UI tasks now run via a JavaScript sandbox so the assistant can steer the UI for you via code. • Retrieval Relevance: Evaluate retrievals from vector search, tools, MCP, web search, or SQL.

100

arize-phoenix @arize-phoenix.bsky.social · 20/08/2026

Did you know that your browser might have an LLM built right into it? With the Prompt API, you can take advantage of local models like developer.chrome.com/docs/ai/pro...). Cool right? Well now you also can get a nice streaming UI to chat with these models, built right into Arize Phoenix.

000

arize-phoenix @arize-phoenix.bsky.social · 04/08/2026

AI Queries and Browser AI Phoenix filters now understand English. Type "responses with apologies" and get a real filter expression back: ▎ 'sorry' in output.value or 'apolog' in output.value

101

arize-phoenix @arize-phoenix.bsky.social · 01/08/2026

Balancing cost, speed, and correctness can be a tricky balance. That's why we need good visuals to figure out the right sweet spot!

000

arize-phoenix @arize-phoenix.bsky.social · 31/07/2026

Customizable visualizations of your agent traces are here. Track cache hits, online eval degradations, tool call errors, all in real-time so you can quickly identify problems in production. What does your agent operations center look like?

100

arize-phoenix @arize-phoenix.bsky.social · 29/07/2026

New Sandbox in Phoenix - Monty (Python)!

000

arize-phoenix @arize-phoenix.bsky.social · 28/07/2026

In an agent session, a cache miss re-bills your entire history at full input price. That's why a "continue" after a coffee break can cost more than the model's actual answer. Cache read vs. write isn't a footnote in your bill. Earendril's post is a must read. earendil.com/posts/promp...

010

arize-phoenix @arize-phoenix.bsky.social · 25/07/2026

Search -> Select -> Download -> Coding Agent. Feed critical traces back to Agents. No fancy words for this one: this is plain old-fashioned debugging. Just now with Agents.

000

arize-phoenix @arize-phoenix.bsky.social · 19/07/2026

phoenix is now is an oauth2 authorization server! `px auth login` opens your browser, you get a short-lived user-scoped token. no long-lived api keys. and admins can audit / revoke every cli and mcp grant from one screen. remote mcp server ships in beta too - more on this soon.

010

arize-phoenix @arize-phoenix.bsky.social · 13/07/2026

Our favorite tools are the ones that have maximum customizability. Last week we added customizable charts, command K, and recent searches. this week we've added custom column ordering. Built to help you have the tables and dashboards you need to monitor agents day in and day out.

010

arize-phoenix @arize-phoenix.bsky.social · 12/07/2026

What does your agent operation center look like?

010

arize-phoenix @arize-phoenix.bsky.social · 11/07/2026

Experiment Baselining and Charts When trying to determine if a new model is up to the task, you need to factor in many dimensions. performance - measured by evals latency - is the model fast enough to give you the right UX tokens - how chatty is the model to achieve the result

110

arize-phoenix @arize-phoenix.bsky.social · 09/07/2026

⌘K is in Phoenix. jump to any project, dataset, prompt, or experiment. No mouse required.

010

arize-phoenix @arize-phoenix.bsky.social · 07/07/2026

Agent traces and trajectories are growing increasingly longer and more complex. We've seen some traces 1000s of spans deep. That's why we've added trace search. Search across a trace and the UI will now show you the call stack to the spans you are looking for across workflows and sub-agent calls.

010

arize-phoenix @arize-phoenix.bsky.social · 07/07/2026

The OSS team has a mantra: 3 clicks to glory. We're not there yet but we're getting closer!

100

arize-phoenix @arize-phoenix.bsky.social · 04/07/2026

Fable 5 might be amazing, but it can't tend the grill 🍔 Happy Fourth of July from our team to yours. We hope you are getting to spend the day with friends, family, and loved ones.

010

arize-phoenix @arize-phoenix.bsky.social · 01/07/2026

PXI now supports sub-agent streaming so you can inspect the execution of the sub-agents you kick off. We've started moving much the skills and tools available in the main agent into the sub-agents to empower interesting delegation patterns.

110

arize-phoenix @arize-phoenix.bsky.social · 30/06/2026

PXI (Phoenix Intelligence) now runs in your terminal! You can now use PXI without leaving your terminal. It's the same agent that powers the in-browser experience, now available as an interactive chat in your shell. npm install -g @arizeai/phoenix-cli@latest > pxi

000

arize-phoenix @arize-phoenix.bsky.social · 24/06/2026

📦🏷️🔖🔑 This week in Phoenix: - server-side bash for PXI subagents, - label management on list pages, - trace-level annotations everywhere - and OAuth2 role-override preservation. arize.com/docs/phoeni...

000

arize-phoenix @arize-phoenix.bsky.social · 17/06/2026

📊 Phoenix 17.7.0 makes your token usage legible. New token detail charts break prompt + completion tokens into their parts, over time: • Prompt → input, cache read, cache write, audio • Completion → output, reasoning, audio

220

arize-phoenix @arize-phoenix.bsky.social · 08/06/2026

Phoenix just hit 10,000 GitHub stars! Three years ago, Phoenix didn't exist. Arize was a closed-source company. A small team was asked to change that. Catch the full interview with the team who made it happen and where AI observability is going next: arize.com/phoenix-10k

010

arize-phoenix @arize-phoenix.bsky.social · 15/05/2026

Something we’ve been playing with and liking a lot: Give every coding agent its own observability stack.

100

arize-phoenix @arize-phoenix.bsky.social · 14/05/2026

Agents are becoming increasingly distributed. A single agentic workflow might cross an app server, an MCP tool server, another agent over A2A, an ACP-compatible runtime, a vector database, an LLM provider, and multiple internal APIs.

A visual representation of a distributed agent workflow with task timelines on the left and a flowchart of interactions on the right.

140

arize-phoenix @arize-phoenix.bsky.social · 09/05/2026

Evals should help you ship FASTER. They are not there to gatekeep . They should be an accelerant: a way to move faster with more confidence. That only works if feedback is fast and accurate. When evals are slow, teams run them less often, test fewer cases, and push quality checks to later.

A diagram illustrating an executor system with parallel workers, rate limiting, and evaluation tasks to optimize throughput over time.

110

arize-phoenix @arize-phoenix.bsky.social · 08/05/2026

The official tanstack AI Otel support is out! Looking for a OSS backend for traces, datasets, and replay? Check out our docs for how to seamlessly add debug-ability. x.com/tan_stack/s...

000

arize-phoenix @arize-phoenix.bsky.social · 08/05/2026

One of the oldest lessons in ML is still one of the most useful for working with LLM apps: Don’t evaluate on the same data used to build. Train/dev/validation/test splits exist for a reason. They help separate “this worked because it was tuned against it” from “this actually generalizes.”

111

arize-phoenix @arize-phoenix.bsky.social · 06/05/2026

Coding agents are growing ~10x year over year. But most are still operating blind. If agents are shipping to production, they need to be grounded in real infrastructure: * Traces → what actually happened * Evaluations → what’s correct vs broken * Feedback → what to fix next

100

arize-phoenix @arize-phoenix.bsky.social · 06/05/2026

Different ways to deploy Arize Phoenix — from local dev to production scale. Phoenix is a lightweight, containerized platform for LLM tracing and evaluation. How you deploy it should match your team size, scale, and reliability needs. Here are the most common patterns:

110

arize-phoenix @arize-phoenix.bsky.social · 05/05/2026

Phoenix 15.4 now supports provider tools A concrete example of why provider tools matter for grounding and tool-call quality.

100

arize-phoenix @arize-phoenix.bsky.social · 05/05/2026

LLM applications are noisy by default. Non-determinism, open-ended inputs, and multi-step tool use create a flood of traces that all look “fine” — until something breaks. The difference between noise and signal is annotation.

110

arize-phoenix @arize-phoenix.bsky.social · 05/05/2026

Most tool-calling evals treat everything as one check. But a tool call actually has 3 separate steps — and each can break independently 👇 🧠 Tool Selection Did the model pick the right tool? ⚙️ Tool Parameters Did it call that tool correctly (valid args, right values)?

100