Riel St Amand

AI Solutions Advocate @getsmartdata
Creator of @mockingboardapp, Kinewright, and proud Métis Canadian

[Riel St Amand @RielStAmand]

@PaulTassi Screams of a startup that will actually go nowhere and just lose a bunch of money

[Riel St Amand @RielStAmand]

Anything to bring Higgsfield down. Horrendous company

[David Im @davidim]

As a builder in the AI video space, I was terrified about how much markup API providers charge for simply routing the models.

Higgsfield API adds a 40% markup even with its maximum advertised discount. Fal adds 100%. Runway adds 414% (Seedance 2 480p)

I might be hated by the industry, but : 


Today, we’re launching Sume API, the first image & video model provider with 0% markup.

For example with Seedance 2.5 1080p
• 50% cheaper than fal
• 50% cheaper than higgsfield api (29% even with biggest discount)
• 28.02% cheaper than kie ai

The routing infrastructure layer is never the moat. We believe that APIs should be commoditized, and the moats will be built at the application layer.

Check the pricing comparisons below, and see how the broken industry is saved.

[Riel St Amand @RielStAmand]

Oh hell yes let’s gooooo. Side note: their launch videos always slap. Them and Anthropic have these things mastered

[Riel St Amand @RielStAmand]

@sambruchhaus @SumerSports I would like to follow up with something about Mondre

[Riel St Amand @RielStAmand]

So far weirdly quiet week in the world of models. Cybersecurity definitely stealing the proverbial show this week

[Riel St Amand @RielStAmand]

@theo Literally people think that GUI = harness

[Riel St Amand @RielStAmand]

I like the name a lot. Partially my multiple decades in piano. Partially it tracks with everything else (Opus, Sonnet in particular)

[Claude @claudeai]

We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.

It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

[Riel St Amand @RielStAmand]

@MiaAI_lab Found the same. I benched it against my own build of Ornith 1.5-35B-A3B (based on the Qwen model) and Ornith was smarter, faster, and used half the tokens to get to answers

[Riel St Amand @RielStAmand]

Still waiting on Claude Duet access. Tragic. Looking at early previews, it seems like it’s more what I might be hoping for in a “personal AI assistant.” Less “manage life and finances and travel” and more “I need you to proactively help me keep track of the 6 different things I manage at work and what needs to happen”

[Riel St Amand @RielStAmand]

Don’t use Max on any Claude model it makes them all worse. I’m not kidding. They literally bench better for far less tokens and cost on xhigh vs max.

[Irushi @Im_IrushiK]

Is Opus 5.5 Max significantly smarter than Opus 5.5 Medium??

Apparently, Opus 5.5 Max handles complex coding tasks much better, while Medium can feel frustrating unless the task is relatively simple.

But Max also takes more time and potentially consumes more usage.

So, Are higher reasoning levels genuinely worth the extra time and tokens, or is Medium enough for most coding tasks?

What reasoning effort do you use as your daily driver???

[Riel St Amand @RielStAmand]

Yeah, guess you could say I’ve made it

[Riel St Amand @RielStAmand]

Seems like Cloudflare and Vercel are the two non-big 3 infra companies to come out of today clearly winning

[Riel St Amand @RielStAmand]

@ryanrhughes @openclaw I’m curious to see how this differentiates itself (if it does). Grok Bot having full access to X and Muse having full access to FB/Instagram were each notable examples of why you might prefer them

[Riel St Amand @RielStAmand]

This is kind of incorrect. 125 tok/s is at 1k context. At more normal context lengths for agentic coding work, this model runs at 60-100 tok/second. Anyway, my M3TiS model runs at 540 tok/s code and fits on just a 3090 with full context.

https://t.co/T9rgVNiIhP

[Richard Socher @RichardSocher]

Flint is also the fastest model in its class:

• 125+ tokens/sec on an M4 Pro MacBook Pro
• 150+ tokens/sec on a 64GB NVIDIA RTX Spark

Up to 2x faster than comparable models on short contexts, and 5x+ faster than dense models with similar coding scores. Your data stays on your computer, and it works offline.

[Riel St Amand @RielStAmand]

Shocking. The fraudster running for governor is also a giant piece of shit! Who could have seen this coming

[FactPost @factpostnews]

Unearthed footage of Vivek Ramaswamy mocking Amy Acton for being a survivor of childhood abuse:

"She has no vision other than to complain about what someone else did to her"

[Riel St Amand @RielStAmand]

Oh great Railway is also hacked. Cool. Why not

[Riel St Amand @RielStAmand]

Joined the waitlist!

[Claude @claudeai]

Claude Duet is an early research preview, available through a waitlist on Pro and Max plans in the US. It can still misjudge what’s urgent, and we’ll keep tuning it based on what early users tell us.

Join the waitlist: https://t.co/kNwzHHeYrO
Read more: https://t.co/GT1xV1RJ7f

[Riel St Amand @RielStAmand]

I know this is a classic "snobby tech bro" take but idc

[Smart Data @getsmartdata]

PHP version 8.5.12, the latest patch, may be itself compromised according to a number of early reports we’re seeing. We once again suggest taking any public facing PHP applications you run offline for the time being.