Sabitlenmiş Tweet
GMI Cloud
1.7K posts
GMI Cloud
@gmi_cloud
AI-native Inference Cloud. Questions: https://t.co/0lGtMMaeY6
Mountain View, CA Katılım Aralık 2023
42 Takip Edilen2.6K Takipçiler
Massive congrats to @lmsysorg and @radixark!
always in awe of your contributions to the inference community
grateful to have you as a partner 🤝
RadixArk@radixark
Today, we are thrilled to officially launch RadixArk with $100M in Seed funding at a $400M valuation. The round was led by @Accel and co-led by @sparkcapital. RadixArk exists to make frontier AI infrastructure open and accessible to everyone. Today, the systems behind the most capable AI models are concentrated in a small number of companies. As a result, most AI teams are forced to rebuild training and inference stacks from scratch, duplicating the same infrastructure work instead of focusing on new models, products, and ideas. RadixArk was founded to change that. We are building an AI platform that makes it easier for teams to train and serve the best models at scale. RadixArk comes from the open-source community. We started with SGLang, where many of us are core developers and maintainers, and expanded our work to Miles for large-scale RL and post-training. We will continue contributing to both projects and working with the community to make them the strongest open-source infrastructure foundations for frontier AI. We would like to thank our long-term partners, contributors, and the broader SGLang community for believing in this mission. We're also grateful to @Accel and @sparkcapital, NVentures (Venture capital arm of @nvidia), Salience Capital, A&E Investment, @HOFCapital, @walden_catalyst, @AMD, LDVP, WTT Fubon Family, @MediaTek, Vocal Ventures, @Sky9Capital and our angel investors @ibab, @LipBuTan1, Hock Tan, @johnschulman2, @soumithchintala, @lilianweng, @oliveur, @Thom_Wolf, @LiamFedus, @robertnishihara, @ericzelikman, @OfficialLoganK, and @multiply_matrix among others. Thanks for the exclusive interview with @MeghanBobrowsky at @WSJ about our vision.
English
Today, we are thrilled to officially launch RadixArk with $100M in Seed funding at a $400M valuation. The round was led by @Accel and co-led by @sparkcapital.
RadixArk exists to make frontier AI infrastructure open and accessible to everyone. Today, the systems behind the most capable AI models are concentrated in a small number of companies. As a result, most AI teams are forced to rebuild training and inference stacks from scratch, duplicating the same infrastructure work instead of focusing on new models, products, and ideas.
RadixArk was founded to change that. We are building an AI platform that makes it easier for teams to train and serve the best models at scale.
RadixArk comes from the open-source community. We started with SGLang, where many of us are core developers and maintainers, and expanded our work to Miles for large-scale RL and post-training. We will continue contributing to both projects and working with the community to make them the strongest open-source infrastructure foundations for frontier AI.
We would like to thank our long-term partners, contributors, and the broader SGLang community for believing in this mission. We're also grateful to @Accel and @sparkcapital, NVentures (Venture capital arm of @nvidia), Salience Capital, A&E Investment, @HOFCapital, @walden_catalyst, @AMD, LDVP, WTT Fubon Family, @MediaTek, Vocal Ventures, @Sky9Capital and our angel investors @ibab, @LipBuTan1, Hock Tan, @johnschulman2, @soumithchintala, @lilianweng, @oliveur, @Thom_Wolf, @LiamFedus, @robertnishihara, @ericzelikman, @OfficialLoganK, and @multiply_matrix among others.
Thanks for the exclusive interview with @MeghanBobrowsky at @WSJ about our vision.
English
@ItsCuthulhu @Kimi_Moonshot they posted a lot of benchmarks for Kimi, some included comparison with other open weight models
English
@gmi_cloud Architecture explains the gap: MiMo's MoE runs more active params per token than Kimi K2.6's optimized routing — hence slowest. DeepSeek V4's 'comprehensive' edge is partly MLA: ~75% KV-cache compression makes it far better for long agentic loops. What task were you testing?
English
@gmi_cloud MiMo might be the slowest but had the best output
English
@bettercallsalva we originally wanna include token burns as well. this is more like speed in completing the task and the visual output for a small game.
will be more specific next time
English
@gmi_cloud Those rankings imply a single eval but the order changes if you measure throughput, params, or task accuracy. Which one was this?
English
@gmi_cloud Impressive benchmark. Kimi K2.6 at 29s for a full playable racer stands out, while GLM 5.1 edges visuals and DeepSeek V4 the feature depth. All four delivering functional games shows solid progress.
English
@ofabdalaX we tested them for a very simple game, the prompt is actually in the video.
this is very helpful to know!
English
@gmi_cloud rodei o mesmo set semana passada e o ranking inverte por task: Kimi K2.6 voa em chat curto mas trava em agentic loop com 3+ tools. DeepSeek V4 mais lento mas hallucina menos em decisao. GLM 5.1 unico que fica consistente em pt-BR. qual era a task?
Português