wkcheng

It would be super useful to have more specific analyses of what areas the model is good at. I'd like to know when I should reach for this model instead of Anthropic's or OpenAI's.

For example, Opus 5.5 is better at UI design than Sol. Anyone who has used both models knows that, but it's not something you can look up anywhere.

I'm sure that there's a need for this kind of benchmarking, not just for coding.

The benchmarks are great for sure.

Now if only they could fix their coding harness so that it's usable for day-to-day use. I paid them for a month to try out Muse Code, but their sandbox never worked across Mac or Linux (both very stock installs, nothing special).

Maybe I'm just not looking in the right place, but I cannot find what the API shape looks like. I've even deployed this model via Foundry and it doesn't say what to POST or what to expect back.

I don't see any API documentation for this yet. How can someone actually try it? Did they rush this out for hype?

When you are running the model locally and trying to do other tasks at the same time, Whisper Large feels too clunky on normal hardware.

I like Parakeet because it's good enough, relatively light, and fairly fast. I'm using this for things like meeting transcription and dictation. Since I'm sending most of the text to an LLM to clean up afterwards, it works well enough.

I'm hoping for something the size of Parakeet (or smaller) but better quality. It feels like with all of the advances in making smaller models better in the LLM space, someone should be able to come up with a lightweight and better quality speech to text model.

How does this compare with Parakeet? I've been using that locally in my projects on an M-series macbook and it's been working great. It's fast and accurate enough for my use cases (meeting transcription, audio transcription for demo videos, etc.)

This definitely seems lighter and faster. How does accuracy compare?

I agree, and I haven't seen other people mention this! The benchmarks for GPT 6 Sol are great, but realistically it does not seem better than 5.6 Sol. 6-Sol is noticeably worse for code reviews (worse than Deepseek 4.1 flash), has implementation issues (requires more rounds of code reviews and fixes to get to a serviceable state). Opus 5.5 is much much better.

I've implemented multiple features side by side with Opus 5.5 and 6 Sol, and the Opus 5.5 results always have fewer high severity bugs and require fewer rounds of fixes to get it over the finish line.

If 6.1 Sol has actually matched Opus 5.5, I'd be very happy. However, benchmarks and real usage don't seem to agree in my own tests. So we'll have to see.

That makes sense. I'm interested in seeing where Haiku 5.5 comes in then when it gets released. It feels like the low intelligence / fast niche will be covered there.

The cost / performance chart shows that in almost all configurations, it looks worse than Opus. Why would you use Sonnet 5.5 on xhigh if you would get better results (higher score, cheaper cost) on Opus 5.5 high?

Is there a good use case? This isn't like Luna where it's much cheaper/effective just to use Luna in certain situations.

Are people using MiMo models as their daily drivers in a company setting? If so, how? I know they're available over OpenCode and directly from Xiaomi, but those are not great options. OpenCode Go straight up doesn't give any guarantees about training on your data, and Xiaomi says that they won't but it's unclear.

With some models you can find hosting companies based in the EU or US, but then you don't know how they're quantizing the models, so you're not sure about the actual output quality.

How are people actually using this? Or are people just experimenting with side projects?

How do people actually use this? Do they use it through some sort of subscription plan, or via OpenRouter?

If the benchmarks are real and reflect actual use, then this is an insane model. This 300B model outperforms the previous DS4 Pro preview model (1.8T params), and it looks like it outperforms GPT 5.6 Luna too. And it's still cheaper than Luna, even with the price decrease.

Crazy.

Just use it through something like Azure. They host the entire model and serve it from the US. I'm sure that there are other providers like this.

We use it that way and it works great.

Use it through Azure! Azure hosts DeepseekV4-Pro and DeepseekV4-Flash themselves. We're using it and it works great.

You don't get the discount that Deepseek is providing, but it's still a cheap model (v4-pro is cheaper than sonnet)

This is nice! I tried it for a bit and it was indeed quite fast. Are you looking for contributors, or are you building this as a personal tool? I ran into some issues when attempting to use different models, though: gpt-5.5 on Azure doesn't work, even with the OpenAI compatible endpoint, because "max_tokens" has been replaced with "max_completion_tokens". And it doesn't appear possible to pass through custom headers, so I wasn't able to specify reasoning_effort for deepseek models.

This was my workflow (this is repro-able 100%): 1. Take an obsidian vault (just a few folders and subfolders of markdown files) 2. Copy the entire folder (keep all of the dates with cp -a) to a new folder 3. Install Tolaria, point it to the new folder you created. 4. Open it up. This creates a git repository. 5. Click on the bottom UI to "Restore Tolaria AI Guidance." After this point, the sorting doesn't work and will continue failing. Everything says that it was created or modified recently.

Edit: I think I found the cause. When you initialize a git repository, the file modified times stay the same, but sorting by modified appears to sort by the commit date, which is now all the same for every file (since it was committed in one go)

Nice work! This looks really cool.

I downloaded and am trying it out, but I'm running into a pretty annoying sorting bug that's preventing me from using it for real. I copied over files from my Obsidian vault (preserving file times), and the first time it loaded, everything seemed to work fine. After doing the first git commit, however, Tolaria cannot seem to sort properly by last modified anymore (I'm getting notes from 2023 or 2025 up at the top). The file system tree still has the correct modified and created times.

Yeah, Parakeet runs great locally on my M1 laptop (through MacWhisper). Transcription speed of recordings feel at least 10x faster than Whisper, and the accuracy is better as well. Push to talk for dictation is pretty seamless since the model is so fast. I've observed no downside to Parakeet if you're speaking English.

Does this support using the Parakeet model locally? I'm a MacWhisper user and I find that Parakeet is way better and faster than Whisper for on-device transcription. I've been using push-to-transcribe with MacWhisper through Parakeet for a while now and it's quite magical.

Yeah, that's the way that I cut onions: you make vertical cuts followed by one single horizontal cut slightly above the cutting board.

This way of calculating doesn't take into account the creative ways you can make cuts. You could also do mostly vertical slices, and then slightly angle inwards when you do the final few cuts. That would get you a more optimal distribution as well.

The attitude of the blog writer in their interactions also feels off. Just reading this blog post makes me think that this person is difficult to work with, requires extremely clear guidelines and instructions, and has a hard time making their own decisions. Maybe this is a good fit for a large, established company, but startups have their own needs.

"Create a terminal inspired email client so we can do an alpha test with some customers" is a reasonable ask for an engineer at an early stage startup. Of course, there would be a bit more specification, but a lot of the details would still be up to the engineer. This applicant wants more certainty than they can get.

This is illustrated by the line: "I would like to know what kind of response I could expect from Kagi if I drive it to completion." This is not a great request to make. There's no way they can answer that question, because there is no certainty available. They're probably getting a few hundred or a few thousand more submissions to evaluate.

Maestro AI | Senior Full Stack Engineer | REMOTE (US) Seattle onsite preferred, Remote OK | https://www.getmaestro.ai/

Hi, I'm William, the co-founder and CEO of Maestro AI. We're building an all-knowing chief of staff for engineering and product leaders. Maestro provides real-time, comprehensive insights, synthesized from Slack conversations, Jira tickets, and more. We allow leaders to stay in control of deadlines, accelerate their team, and understand and prioritize work.

Our goal is to eliminate information silos and become the OS that helps run engineering teams (and later every type of team).

Here's a recent demo video: https://www.youtube.com/watch?v=Xd7_0fURLxU

We're an early-stage, venture-backed startup, and you'll have the opportunity to shape our product and culture from the ground up.

We're looking for someone who takes pride in their craft, pushes their limits, and takes action to create something new. Our stack is Python on the backend and React/Redux/Typescript on the frontend. We're training our own models in addition to using existing LLMs.

To apply: Please email will+hn at getmaestro.ai with your resume, your Github, and a short note detailing a project you've enjoyed working on. View a full list of our roles here: https://maestroai.notion.site/Maestro-AI-Careers-51385f9fca9...

We've been thinking about the things you need to be careful about when building AI products for businesses and put them in this article. Are there important things that we missed?

It's insane that that this works, and that it works fast enough to render at 20 fps. It seems like they almost made a cross between a diffusion model and an RNN, since they had to encode the previous frames and actions and feed it into the model at each step.

Abstractly, it's like the model is dreaming of a game that it played a lot of, and real time inputs just change the state of the dream. It makes me wonder if humans are just next moment prediction machines, with just a little bit more memory built in.

Maestro AI | Senior Full Stack Engineer, Senior AI Engineer | REMOTE (US) Seattle onsite preferred, Remote OK | https://www.getmaestro.ai/

Hi, I'm William, the co-founder and CEO of Maestro AI. Throughout my career in software engineering, I've repeatedly witnessed how fragmented workflows and broken processes can stifle even the most brilliant teams. That's why I started Maestro AI. Our mission is to streamline the process of software development, turning chaos into clarity so that teams can accomplish more.

Maestro AI provides real-time, comprehensive insights for teams so everyone can easily see what's happening, how projects are progressing, and where people are blocked. We eliminate information silos by aggregating data from multiple collaborative tools like Slack, Jira, Notion, and Github, enabling seamless communication and informed decision-making across the entire organization.

Here's a recent demo video: https://www.youtube.com/watch?v=Xd7_0fURLxU

We're an early-stage, venture-backed startup, and you'll have the opportunity to shape our product and culture from the ground up.

We're looking for someone who takes pride in their craft, pushes their limits, and takes action to create something new. Our stack is Python on the backend and React/Redux/Typescript on the frontend. We're training our own models in addition to using existing LLMs.

To apply: Please email careers+hn at getmaestro.ai with your resume, your Github, and a short note detailing a project you've enjoyed working on. View a full list of our roles here: https://maestroai.notion.site/Maestro-AI-Careers-51385f9fca9...

Does anyone know whether the 128K is input tokens only? There are a lot of models that have a large context window for input but a small output context. If this actually has 128k tokens shared between input and output, that would be a game changer.

This may be why things like Substack and Beehiiv have taken off. The only way to combat reposting through Google is to deliver content directly to email inboxes before it gets ranked and reposted.

There is something additional at play with TechCrunch, though. Recently I feel like they haven't been posting as many articles that are about smaller startups as they used to. They tend to post more about Google, Nvidia, Intel, etc. I find myself reading it less and less because it's mostly news that you can also find elsewhere.

Yeah, you're right. Data breaches are essentially just slaps on the wrist to companies like AT&T. Maybe it's possible to fine them based on the proportion of the userbase that was affected and the profits they generated for a certain time period.

I wonder if this will push companies to stop using external vendors to store and process data. If companies stored all of their info in house, it would prevent the case where compromising one vendor compromises everyone's data. But it would also mean that each individual company needs to do a good job securing their data, which seems like a tall ask.

I wonder if allergies can eventually be fixed somehow as well. There has to be an immune or autoimmune reason why some people have tons of allergies and other people are perfectly fine.