Morph

making coding agents better with specialized inference
WarpGrep:

[Morph @morphllm]

Introducing FlashCompact - the first specialized model for context compaction

33k tokens/sec
200k → 50k in ~1.5s
Fast, high quality compaction

[Morph @morphllm]

@katedeyneka openai, anthropic, gemini, deepseek

[Morph @morphllm]

Introducing the Morph Model Router.

It chooses the best model for each task in under 50ms.

Keep frontier performance while lowering latency and cost. Available today in our API

[Morph @morphllm]

Being model agnostic is better, faster, and cheaper

Try the Morph Router API:
https://t.co/X0glPN6rgn
or dm us to self host

[Morph @morphllm]

Not every agent step needs your most expensive model.

It’s not “cheap vs smart.”
There are many points on the cost-performance curve.

Model intelligence is jagged. Routing lets you use the right model for the right task, improving quality while cutting cost 25-50%

[Morph @morphllm]

@latentlocal every request gets the raw output of the classes so you can do whatever you'd like with it

[Morph @morphllm]

Morph AutoRouter is now 5x faster

know difficulty, ambiguity, and domain in under 50ms

[Morph @morphllm]

vibe test it here
https://t.co/r3X1WnfXM8
integrate here
https://t.co/X0glPN6rgn

[Morph @morphllm]

Why these classes?
We think they're useful for routing between GPT vs Opus vs open-source

how 30ms?
megakernel

[Tejas Bhakta @tejasybhakta]

the general applied standard intelligence compute company

[AI Engineer: Miami @AIEMiami]

Join us in welcoming @morphllm Founder, @tejasybhakta to the AIE Miami lineup!

Don't miss his talk 'Everything is Models' next week on the big stage!

Get your tickets: https://t.co/JMbrvl07Ah

[𝗥𝗬𝗔𝗡 𝗟𝗘𝗘 @ryanleecode]

warpgrep_github_search from @morphllm is probably the most unfathomoly unfair advantage you can have right now. 10x better than grep app

Even beats Ctx7 tbh

https://t.co/WpQF6W63RW

[Morph @morphllm]

@ryanleecode @kunchenguid @dhruvbhatia0 is working on a fix for the latest CC version, but we reccomend compacting at around 150k for best coding performance. for investigation, search, chat with repo type uses, no need to compact until 1m

[Morph @morphllm]

@tube_you86806 @FactoryAI oh wait actually no it doesnt. @FactoryAI wanna expose a compactions hook?

[Morph @morphllm]

Introducing FlashCompact - the first specialized model for context compaction

33k tokens/sec
200k → 50k in ~1.5s
Fast, high quality compaction

[Morph @morphllm]

@tube_you86806 @FactoryAI yep! https://t.co/hGUHm9179H

[Morph @morphllm]

@steipete @Techguy0101010 @openclaw @steipete good to go? https://t.co/A24Tv43XLe

[Morph @morphllm]

https://t.co/DXpHGbkwzS
install the plugin + mcp here

[Morph @morphllm]

Our Claude Code plugin is here!
- WarpGrep for state of the art fast code search
- FlashCompact, our specialized fast compaction model

end to end speedup on long claude code sessions: -37%, while saving claude tokens and improving accuracy

[Morph @morphllm]

@fastquant0 clerk outage with their bot protection feature.
@clerk please fix

[Fondo.com @Fondocom]

Agents don’t need bigger models. They need better tools.

Morph trains coding subagents.
Not for humans. For frontier models.

Fast Apply edits at 10,000 tokens/sec.
WarpGrep handles code and log search.

Both keep the main model’s context clean

Because when context gets too large, performance drops.

Now Morph is pushing coding subagents even faster

One newer model runs at 33,000 tokens/sec: https://t.co/8P38gPhxur

🎙️ @tejasybhakta, Founder & CEO, @morphllm on @fondocom @thestartpod w/ @davj