arizen

Seems like learning rate velocty may be the ultimate moat

Does it have /goal feature similar to Codex?

Simple as it sounds but I ask Sol or Astra to propose optimal model and effort level for a given workload. Works pretty neat for me.

Astra incoming!

Is there any good subscription and CLI harness to use Gemini models now?

I tested Gemini CLI while ago, and it was awful tbh.

The only company to use Claude.md instead of Agents.md standard

Partially frontier moved to cost and speed axes.

There are also markets with scarce supply, like desired properties in prime locations. Probably some other similarly structured markets, as well.

So there is also a question, what would be happening to these.

Probably apples to apples would be to compare z.ai subscription plans vs API pricing

Few weeks ago, I wouldn't expect this statement to be true. Accelerate!

Why not also Qwen?

Could these complex/hard to read Fable outputs be sign of some kind of industrial level of intelligence, which us humans may have a hard to comprehend, while it may be also hard for machine to use simpler texts to properly outline all nuances and complexities of concepts it output?

Have you tested Qwen3.8-Max yet?

Probably increasing learning rate have some value for these players which is enabled by subsidized usage.

And cost of acquiring learning via users feedback versus additional learning data value for a given company will dictate whether such company will give up or will continue subsidized usage, despite being unprofitable on the paper, but probably really valuable as a long term strategy for GTM and product development and increased learning rates.

And DeepSeek V4 Flash + GLM 5.2 is a really good blend of both (fast/cheap DS + more intelligent GLM)

Yes, true creativity usually or mostly comes from real constraints, in my experience.

As, if there are no constraints in some specific area, there is no kinda "survival need" to improve there, hence brain is not working as hard/smart/deep as it could.

"GPT‑5.6 delivers a step change in design judgment. With only high-level direction, GPT‑5.6 creates tasteful, ergonomic, and functional interfaces. Its stronger computer-use capabilities let it inspect and refine the rendered result—not just generate the underlying code or content—so it can catch visual and functional issues and apply finishing touches before handing the work back."

This one is really promising, as it may allow to close major gap with Claude in design/UI skills

I'm still confused is it available to public via some sort of subscription?

It was not idiotic from capitalism short-term profit gains point of view.

Sol Goodman

Ditto on GLM 5.2 + DeepSeek V4 Flash combo.

For most important work (complex, cross-domain inquiries etc.), I still rely on Codex GPT 5.5 though.

Some anecdata on Fusion: I run same query I used for Fable on OR Fusion and results were worse.

It felt, like Fable was able to kinda grasp very deep knowledge/intelligence layers and outline solution not only in agreeable way, but rather it proposed to prioritize solution items, with discarding some of the items, which made a lot of sense to me.

While Fusion felt more like a bit diversified answer of the same class of pre-Fable SOTA models, without touching the depth of knowledge/intelligence layers, which Fable was able to get, in my very limited tests I did, while Fable was accessible.

OpenCode Go is pretty good in my experience too.

I ended up using DeepSeek V4 Flash as main workload model, while keeping DeepSeek V4 Pro and Qwen 3.7 Plus as advisors on system architecture and other advanced matters to guide DS Flash.

I run a simple benchmark on OpenCode Go models while ago, if anyone want to read more: https://arizenai.com/seven-models-judged-each-other/

opensourcing software may enable leverage of wider network of contributors to given piece of software,hence software can evolve much more quickly and efficiently.

Probably there will always be frontier surface which frontier model of a given generation would not be able to automate.

Curious how it compares to Browser Use

Happiness is a derivative of purpose. If someone optimizes their life strictly for happiness while deprioritizing purpose, they likely won't achieve either.

Pursuing a meaningful goal almost always requires enduring unpleasant phases and friction along the way.

Out of curiosity, what specific use cases on programmatic SEO are you currently doing with Opus?

"Don't believe everything you read on the Internet"

-Isaac Newton

Hi HN,

I wanted to share a reference implementation I architected for moving AI Agents from local prototypes to production services.

The Context:

It is relatively easy to get an agent working on a local machine where you can watch the terminal output and restart it if it gets stuck. However, the architecture often breaks down when moving to a headless, hosted environment where the agent needs to handle loops, persistent state, and structured output failures autonomously.

The Solution:

This repo is a 10-lesson lab where you build an "AI Codebase Analyst" designed to handle those operational constraints.

Key Architectural Decisions:

1) State Management (LangGraph): We use LangGraph to implement the State Machine pattern rather than a linear Chain. This provides a standardized way to handle cyclic logic (loops) and persistence without writing "spaghetti code" while loops.

2) Reliability (Pydantic): Treating the LLM as a probabilistic component. We wrap tool calls in strict Pydantic schemas to catch and retry malformed JSON before it hits the application logic.

3) Deployment (Docker): A production-ready Dockerfile setup for serverless environments.

The Repo Structure:

starter branch: A clean boilerplate to build from scratch.

main branch: The full solution code.

curriculum/ folder: The step-by-step guide.

Happy to answer questions about the stack or the trade-offs involved.