Swayam

AI Infrastructure Architect | Multi-Agent Systems | Writing Superagency

[Swayam @0xsuperagent]

Buzz by @blocks is most critical infra app to enable organizations go ai-native bottom-up.

and it is opensource🫰

[Swayam @0xsuperagent]

@AnthropicAI anthropic sounds desparate

[Swayam @0xsuperagent]

K3 goes opensource

[Kimi.ai @Kimi_Moonshot]

Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.

Model weights: https://t.co/7m7eEg6Y0B
Tech report: https://t.co/yeu6cjpMCT
Tech blog: https://t.co/YTfiMSNM1f

[Swayam @0xsuperagent]

Trying to ban open source is incredibly concerning.

Bitchat is critical for such times. Set it up!

[calle @callebtc]

UPDATE: Bitchat is now mirrored on @Radicle — censorship-resistant, peer-to-peer code hosting:

iOS/macOS: rad:z2v9tRJz1oknFAqCSY5W5c76nVvm6
Android: rad:z347HbE1F4c77DVLvsPe2ApoEEw6f

Clone it. Seed it. Keep it alive. 👇

[Swayam @0xsuperagent]

Agentic AI is just a form of control systems.

[Swayam @0xsuperagent]

@vicharabhio 🤣🤣

[Swayam @0xsuperagent]

Kimi K3 just dropped, and the stats are worth sitting with.

Here's what it did 👇
→ Became the world's first open 3T-parameter model. The biggest of its kind ever released to the public.
→ 1M context window. Multimodal.

→ Writes code at the frontier, topping the hardest engineering benchmarks in its class. [Scored more than Fable 5 on frontend]
→ Designed a working chip from scratch that sustains 8,700/tps.
→ Compressed a two-week astrophysics project into two hours. 20+ papers read, 300+ equations checked.
→ Traced 42 years of semiconductor history across 2,800 web searches, unsupervised.
→ Analyzed 391 gravitational-wave events using 20 agents working in parallel.
→ Scored 93.5 on GPQA-Diamond, a PhD-level science exam.
→ Ships cheaper than you'd expect, thanks to a 90%+ cache-hit rate on real coding work.

Full weights release July 27, 2026.

An open model is now doing work that felt closed-lab-only a year ago.
The game is shifting.

[Swayam @0xsuperagent]

@BraedendotTECH @NotIlanCohen Continuous comprehensions are exhaustive.

[Swayam @0xsuperagent]

@thekitze us

[Swayam @0xsuperagent]

@thekitze Before January 2026, I was completely hooked on playing padel.

Now, I’m mostly busy running LLMs
(playing padel only a few times a week).

It’s the quick feedback loops that make this so addictive!

[Swayam @0xsuperagent]

ChatGPT gobbled up Lovable and Vercel.

[Swayam @0xsuperagent]

@boxmining not yet (netherlands)

[Swayam @0xsuperagent]

@kimmonismus I just tried the new version, and the live translation isn’t quite working as expected. It seems to get mixed up between languages sometimes.

However, the turn-taking is much improved compared to before.

Gemini Live is still the best in its class right now.

[Swayam @0xsuperagent]

Official kickoff for agent economy

[Cloudflare @Cloudflare]

We're opening the waitlist for our Monetization Gateway, which will allow you to charge for any web page, dataset, API, or MCP tool behind Cloudflare. The charges will settle in stablecoins over the x402 open protocol. https://t.co/pvICtEIixj

[Swayam @0xsuperagent]

@benln In!

[Swayam @0xsuperagent]

@thsottiaux visualizations

[Swayam @0xsuperagent]

@juliarturc Visual learning and reforcing concepts in different forms.

[Swayam @0xsuperagent]

@DaveShapi Gemini Flash 3.5 is fast but tends to extend simple tasks (and consumes more tokens). Internal orchestration could further be optimized.

[Swayam @0xsuperagent]

@thsottiaux This is just tool description which we can edit now?