Lakshminath Dondeti

CoFounder of FinalLayer (and Rizzle)

[Lakshminath Dondeti @ldondeti]

@grok Wow! Better than Nano Banana in @arena benchmark on text to image. I’m going to try this out today!

[Lakshminath Dondeti @ldondeti]

@arena @SpaceXAI Wow Grok Imagine beating Nano Banana! I thought their text fidelity is not that great!

I'm going to try this out today!

[Lakshminath Dondeti @ldondeti]

@ClaudeDevs Good stuff. I need this. Sessions messaging each other actively would speed things up and avoid the need for rollbacks.

[Lakshminath Dondeti @ldondeti]

@gdb I hope Astra is available widely and not limited like Mythos!

[Lakshminath Dondeti @ldondeti]

@bcherny Yeah. Permission prompts are annoying anyway! Auto mode FTW. 🙏🏾

[Lakshminath Dondeti @ldondeti]

@StateDept Future proofing. That's good!

[Lakshminath Dondeti @ldondeti]

@OpenAI Astra is your Mythos?

[Lakshminath Dondeti @ldondeti]

@elonmusk I guess AI is creating jobs after all!

[Lakshminath Dondeti @ldondeti]

@elonmusk Need a Mac app like Claude Code and Codex apps.

[Lakshminath Dondeti @ldondeti]

@ClaudeDevs Good! Auto mode makes sense anyway!

[Lakshminath Dondeti @ldondeti]

@elonmusk Absolutely. It was a bad idea to remove testing! There are 2 types of tests. And the tests anyway are just one of dozens of criteria.

[Lakshminath Dondeti @ldondeti]

@OpenRouter @OpenAI It being GPT 5.6 Luna!

[Lakshminath Dondeti @ldondeti]

@OpenRouter Yep. It's going to be a popular model. Isn't @OpenAI giving it away practically?

[Lakshminath Dondeti @ldondeti]

Where are the GPT models in this chart @arena?

That out of the way, kudos to @AIatMeta for making it to the top of this chart!

[Arena.ai @arena]

Muse Spark 1.2 (xHigh) is a significant improvement from Muse Spark 1.1 (#11 -> #4). By category, it’s strong in Multi-Turn (#8) and Hard Prompts (#13).

[Lakshminath Dondeti @ldondeti]

@elonmusk I guess I have to feel that religious experience at some point!

[Lakshminath Dondeti @ldondeti]

@byHeatherLong Seems like bad news all the way around.

Small businesses like ours are likely to be careful in increasing spending or hiring after seeing this. That of course makes things worse! ☹️

[Lakshminath Dondeti @ldondeti]

At the very top is once again @cerebras delivering 889 TPS with GPT-OSS-120B.

[Lakshminath Dondeti @ldondeti]

I hunt for models with highest tokens per second on @OpenRouter

Coming in at 149 TPS is GLM 5.2 from https://t.co/ybGrFGNmUn

[Lakshminath Dondeti @ldondeti]

Next up, @cerebras at rocket speed of 477 TPS with GLM 4.7 from https://t.co/ybGrFGNmUn

[Lakshminath Dondeti @ldondeti]

@amiruci @huggingface 185 TPS! 🙌