Introducing FlashCompact - the first specialized model for context compaction 33k tokens/sec 200k → 50k in ~1.5s Fast, high quality compaction

Morph now has the highest cache hit rate out of any provider serving Kimi K3 $200 spent on morph gets you 15% more Kimi tokens than that same $200 would on Modal and 64% more than it would on baseten will be doing the same for GLM-5.3 next

the one-man inference provider i hit a $6m run rate as a solo dev morph was 1 person until last week the first inference provider to be live with kimi k3 without early weights access 250 customer slack connect channels i could've made morph a 1-person $1b company, but it would've been for the Ego instead of whats optimal we are no longer a 1-person company. but we will be the first 10-person $10b company. everyone joining has had multiple offers from frontier labs if you’re top 0.0001% at having adhd, dm me

Morph 🤝 SGLang We're working with the @sgl_project and @lmsysorg team to push open-model disaggregated inference faster together. First up is Kimi K3 Fast, served at up to 100 tokens per second through Morph's OpenAI and Anthropic-compatible APIs. This is the beginning of a deeper collaboration around low-level optimization and high-performance open-model serving. More soon.

fixed it

Kimi performs comparably with Fable 5 on all benchmarks, and outperforms on web dev Integrate: run it in the background on prod with 1 line: MORPH_API_KEY= YOUR_APIKEY npx -y @morphllm/morph-setup --kimi

Kimi K3 comes to Morph via API and OpenRouter. tomorrow 8am PST.

morph has signed the open weights letter open weights matter because the gradient of capitalism corrupts absolutely in the absence of fair competition thanks to @Microsoft for starting this and @JensenHuang for spreading it

@morphllm has signed the open weights letter the models train on the inheritance of humanity. humanity should remain capable of possessing, understanding, and extending them. the deepest argument for open weights is not that open models are cheaper, or even that they create more competition. it is that models are compressed civilization. they train on the accumulated work of scientists, programmers, writers, artists, institutions, open-source communities, and generations of people who made their knowledge legible to others. model developers add enormous amounts of engineering, compute, judgment, and risk to this inheritance. that work is difficult and deserves to be rewarded. but it would be a strange equilibrium if models could learn from the open output of an entire civilization while the resulting intelligence remained available only through a small number of closed interfaces. open weights turn intelligence from a service you receive into a material you can work with. you can inspect it, modify it, specialize it, run it on your own infrastructure, and preserve what you build on top of it. a hospital, factory, university, startup, or country can accumulate capability without depending on the permission of a single provider. this is how real technological ecosystems form. not around one perfect oracle, but around thousands of people adapting a shared foundation to problems its original creators could never have anticipated. there are real risks to releasing powerful models, and openness is not automatically safe. but concentration is not automatically safe either. a future built around a handful of closed systems is fragile in ways we are only beginning to understand.

morph is now a top 30 provider on @OpenRouter our competition has raised over 100x what we have

use models like GLM 5.2 through morph inside of kilocode now

we also built it to fall back to cheaper models when a task doesn't need the expensive ones, thanks to @morphllm routing. keeps it fast and affordable without losing quality.

Excited to be working with Merge to bring GLM 5.2, deepseek v4 flash and more to their gateway!