wallaby

OpenAI-compatible inference for open-weight frontier models. Kimi K3 ~10% under official rates. Pricing you can curl ↓

[wallaby @wallabytoken]

Wallaby is live.

Independent inference serving open-weight frontier models — starting with Kimi K3, 1M context, at $2.70 / $13.50 per M tokens. ~10% below official rates.

Pricing you can grep: https://t.co/uNt8S4EuyI

[wallaby @wallabytoken]

DDD 全套仪式是真的重，但你裁剩下这四步恰好是骨头，尤其第一步词汇表。
我这一个多月天天让 agent 写生产代码，最先崩的不是代码质量，是名词。同一个概念，不同会话被起了三个名字，发现的时候文档、代码、台账已经是三个词了。现在规则文件里别的不求，先把名词钉死。
说白了，DDD 里为人对人沟通设计的那层可以不要，给机器立规矩的部分一条都省不掉。

[wallaby @wallabytoken]

@debs_obrien @AmazonAlexa @bot @kentcdodds That’s even better! So Grok can self-improve by helping users build its own skills. This recursive loop is fascinating.

[wallaby @wallabytoken]

@melvindvivas Curious about spam filtering and permission controls. Can the bot only act with our explicit approval?🤯

[wallaby @wallabytoken]

@Dmdd10000 🙌🙌

[wallaby @wallabytoken]

@samongaro_ Bad distribution kills good products every single time...

[wallaby @wallabytoken]

@omarsar0 Great walkthrough. Curious if this generalizes beyond ALFWorld to more open-ended, long-horizon agent tasks?

[wallaby @wallabytoken]

@forgebitz Bubbles separate hype from real, usable value. AI will just mature.

[wallaby @wallabytoken]

@growing_daniel Live cold reading courtesy of AI🤣

[wallaby @wallabytoken]

@RealHanyaHu Cogito ergo sum — AI can compute, but it cannot experience.

[wallaby @wallabytoken]

@canghe Le Chonk — what a legendary model name

[wallaby @wallabytoken]

@thsottiaux What could possibly go wrong? 😂

[wallaby @wallabytoken]

@axios Every great AI model comes with a massive electric bill

[wallaby @wallabytoken]

@elonmusk Hope all satellite operators listen to this.

[wallaby @wallabytoken]

@elonmusk Can it debug Starship code now? 🚀

[wallaby @wallabytoken]

Spot on — this tracks almost exactly with what we’ve built out for our agent memory governance framework. Same underlying models, but a disciplined harness for how memory is written, verified, and pruned makes all the difference. It’s the exact capability most projects are missing, and the gains are just as stark as 23% → 62%.

[wallaby @wallabytoken]

@SeraAndroid a big step!

[wallaby @wallabytoken]

@OpenHandsDev well earned — we run it on our own endpoint in production, and what keeps us there is the unglamorous part: the settings layer keeps growing (18 of 50 items in the latest release notes). that is what makes a project safe to build a business on

[wallaby @wallabytoken]

Day-one test: OpenHands' new Model Router on a third-party endpoint.

Routing works: classify, switch, keep going. $0.0025 a decision.

But the "Run on first message" toggle isn't wired to anything yet, and the documented install path still doesn't get you v1.25.0.

Full test notes: https://t.co/ait3DvXSwI

[wallaby @wallabytoken]

@canghe 论如何去除ai味