Twitxy
Saved
@ronaldmannak's profile picture
View on X

Ronald Mannak

Verified account This account is verified. Learn more

@ronaldmannak

Local AI. Building @picogpt
Download: apps.apple.com/us/app/pico-pr…
Discord: discord.gg/Nrf5y8Uaxw
San Francisco, CA picogpt.app Joined April 2008
6.5K Following 6.9K Followers 29K Posts

Latest posts 5

Pinned post

Ronald Mannak Verified account @ronaldmannak
Apple Silicon + Gemma 4 fans: this is for you.

Pico AI Server now supports continuous batching with MLX-Swift.

43 tok/s on 1 stream.
26 tok/s per stream on 2 concurrent streams.

That’s 52 tok/s total. a 21% throughput gain on a six-year-old MacBook Pro M1 Max!
1:42
14 replies 23 reposts 363 likes 55K views View on X
Ronald Mannak Verified account @ronaldmannak
Interesting. Also good to see some iOS OGs in the discussion
Ivan Tokar Verified account @ivatokar 16h
An interesting Swift Evolution pitch discussion asks a big question: what would it take to write GPU code in Swift?

The thread started with an experiment translating Swift into Metal. The discussion on November 10 turned toward a more practical pain point: using Metal shaders with Swift Package Manager.

One contributor shared a SwiftPM compiler-plugin workaround and argued that first-class Metal support would help. Others questioned whether a new shader language would solve the bigger problems of GPU programming.

This is an open community discussion, not an accepted Swift proposal or a shipped compiler feature. Still, the tooling question is worth watching.

#SwiftLang #Metal
forums.swift.org/t/swift-on-the…
View quoted post on X
Ronald Mannak Verified account @ronaldmannak
A must have for swift devs
Mattt Verified account @mattt 17h
Over the last 30 days, Sosumi.ai has served tens of millions of requests to coding agents looking up Apple docs, from about 6,900 (nice) unique IPs/day.

But does it actually help? I built an eval to find out: 41 Swift tasks, mostly new APIs in the OS 27 SDKs, checked by the compiler and per-task rules.

Without docs, Opus, GPT-6.1 Sol, and Sonnet pass about a third of them. With Sosumi, they pass 98%, 97%, and 94%.

Compared to built-in web search, Claude models score 12 to 30 points higher with Sosumi, at about half the cost or less. GPT does slightly better with web search (99%), but at 3× the cost.

Running the eval also turned up a handful of fixes, now live (one framework page was 3 MB 😬). The eval is open source: github.com/NSHipster/sosu…
Two bar charts from the Swift API eval: 41 Swift tasks, mean of 3 runs per task. Each model has three conditions: no docs, built-in web search, and Sosumi. Tasks passed: Claude Opus 5.5, 38%, 84%, and 98%. GPT-6.1 Sol, 33%, 99%, and 97%. Claude Sonnet 5.5, 31%, 63%, and 94%. Claude Haiku 5.5, 23%, 64%, and 76%. API cost per task, in US cents: Claude Opus 5.5, 2.5, 11.2, and 5.7. GPT-6.1 Sol, 1.5, 4.0, and 1.3. Claude Sonnet 5.5, 0.8, 4.5, and 2.2. Claude Haiku 5.5, 0.1, 2.0, and 0.2.
View quoted post on X
Ronald Mannak Verified account @ronaldmannak
👀
Qwen Verified account @Alibaba_Qwen 1d
Qwen4 is coming.

Deeper knowledge. Richer world understanding. Stronger coding and agentic capabilities.

Built for real-world tasks — from complex coding and knowledge work to front-end builds and professional software workflows.

Qwen4-Max is now in preview on Alibaba’s Token Plan. You can also get early access to the new model on QwenWork, Qoder, Qwen APP, and QwenStudio.

Put it to the test on a real problem. Show us what you build. 👀

Token Plan: qwencloud.com/pricing/token-…
View quoted post on X