Pinned Tweet
Over the past few months, we’ve worked closely with frontier open-source model teams worldwide on Day 0 support and inference optimization in TokenSpeed.
We also collaborate on speculative decoding—from algorithm research and training draft models with TorchSpec to efficient .
Efficient inference is essential to making AI accessible at global scale, and we believe the infrastructure behind it must be open.
github.com/lightseekorg/t…
We also collaborate on speculative decoding—from algorithm research and training draft models with TorchSpec to efficient .
Efficient inference is essential to making AI accessible at global scale, and we believe the infrastructure behind it must be open.
github.com/lightseekorg/t…
5
4
20
10,809
Cat using Kimi K3.1 by Kimi K3.1
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1…
0
1
10
1,219
LightSeek Foundation retweeted
Thanks to @lightseekorg for Day-0 support with TokenSpeed!
Day-0 support for @MiniMax_AI M3.1 Flash is live in TokenSpeed.
DSpark speculative decoding and MXFP8 + NVFP4 mixed precision on Blackwell.
TokenSpeed is featured in the official deployment guide.
📖 huggingface.co/MiniMaxAI/Mini…
👉 github.com/lightseekorg/t…
11
2
190
20,401
Day-0 support for @MiniMax_AI M3.1 Flash is live in TokenSpeed.
DSpark speculative decoding and MXFP8 + NVFP4 mixed precision on Blackwell.
TokenSpeed is featured in the official deployment guide.
📖 huggingface.co/MiniMaxAI/Mini…
👉 github.com/lightseekorg/t…
DSpark speculative decoding and MXFP8 + NVFP4 mixed precision on Blackwell.
TokenSpeed is featured in the official deployment guide.
📖 huggingface.co/MiniMaxAI/Mini…
👉 github.com/lightseekorg/t…
5
2
17
9,602
LightSeek Foundation 🤗 PyTorch Foundation
We welcome LightSeek Foundation @lightseekorg as a Silver Member of PyTorch Foundation.
Its membership deepens LightSeek Foundation’s commitment to building open inference infrastructure with the community. LightSeek Foundation’s TokenSpeed and SMG are PyTorch Ecosystem projects.
Read the announcement: pytorch.org/blog/pytorch-f…
0
0
6
518
LightSeek Foundation has joined the @PyTorch Foundation as a Silver Member.
TokenSpeed and SMG are both PyTorch Ecosystem projects. This membership deepens our commitment to building open inference infrastructure with the community.
As our president @zhyncs42 shared: “Efficient to making AI accessible at global scale, and we believe the infrastructure behind it must be open.”
We look forward to working with the PyTorch community to improve performance, lower costs, and make AI more accessible.
Read the announcement: pytorch.org/blog/pytorch-f…
TokenSpeed and SMG are both PyTorch Ecosystem projects. This membership deepens our commitment to building open inference infrastructure with the community.
As our president @zhyncs42 shared: “Efficient to making AI accessible at global scale, and we believe the infrastructure behind it must be open.”
We look forward to working with the PyTorch community to improve performance, lower costs, and make AI more accessible.
Read the announcement: pytorch.org/blog/pytorch-f…
0
2
10
35,583
EAGLE-4 is here—a game changer for agentic coding, with massive decoding speedups.
Honored to be a day-0 engine partner with @EagleCorp
Train with TorchSpec. Serve with TokenSpeed. Training and inference in one diagram 👇
Technical deep dive: lightseek.org/blog/eagle-4
Honored to be a day-0 engine partner with @EagleCorp
Train with TorchSpec. Serve with TokenSpeed. Training and inference in one diagram 👇
Technical deep dive: lightseek.org/blog/eagle-4
35
34
195
45,659
LightSeek Foundation retweeted
EAGLE-4 is here.
From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.
EAGLE-4 is our fastest yet.
961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @nvidia B300, TP4, one request, T=0).
Github: github.com/eagle-hq/EAGLE…
From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.
EAGLE-4 is our fastest yet.
961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @nvidia B300, TP4, one request, T=0).
Github: github.com/eagle-hq/EAGLE…
911
743
11,727
5,202,364
DeepSeek @deepseek_ai has just open-sourced its LLM inference engine. TokenSpeed will learn from and adopt these optimizations. Stay tuned!
👉 github.com/deepseek-ai/De…
⭐️ github.com/lightseekorg/t…
👉 github.com/deepseek-ai/De…
⭐️ github.com/lightseekorg/t…
2
2
69
9,569
TokenSpeed Kernel at today’s Triton Conference poster session!
Portable APIs and high-performance kernels for multi-silicon LLM inference. The AMD backend is fully rewritten in Triton + Gluon for ease of use and speed. 👇
⭐️ github.com/lightseekorg/t…
Portable APIs and high-performance kernels for multi-silicon LLM inference. The AMD backend is fully rewritten in Triton + Gluon for ease of use and speed. 👇
⭐️ github.com/lightseekorg/t…
4
6
37
8,171
Congratulations on the launch. We’re honored to be the open source engine partner, and TokenSpeed will support Minimax M3.1 Flash from day one. Stay tuned!
MiniMax M3.1 Flash: Space Performance + Bunny Speed
🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash.
💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence s Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed.
API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens.
Model weights will arrive on Oct 23.
API: platform.minimax.io
M Plan: minimax.io/m-plan
MiniMax Code for desktop: agent.minimax.io/download
0
0
15
1,417
🔹 Day-0 support for DeepSeek V4.1 Flash 1017 in TokenSpeed. Out of the box on Hopper, Blackwell, Grace Blackwell & AMD MI355X.
🔹 TokenSpeed runs reliably in production across thousands of Hopper GPUs.
🔹 Vera Rubin & AMD MI455X optimizations are underway.
👉 github.com/lightseekorg/t…
🔹 TokenSpeed runs reliably in production across thousands of Hopper GPUs.
🔹 Vera Rubin & AMD MI455X optimizations are underway.
👉 github.com/lightseekorg/t…
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, grade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
4
2
23
12,487
We’re thrilled to be Reflection AI’s open ecosystem partner! 🤗 TokenSpeed is one of the deployment solutions. 👇
developers.reflection.ai/deployment-ove…
developers.reflection.ai/deployment-ove…
Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners.
We're releasing:
- Weights, tokenizer, and reference implementation
- BF16, FP8 and NVFP4 weights and a DFlash2 drafter
- The open source ained Beam on: Mirror CLI
Weights and model card on Hugging Face: huggingface.co/reflection
2
1
19
721
Reflection @reflection_ai 's Beam, day 0 on TokenSpeed.
8× NVIDIA GB200 · NVFP4 + DFlash 2
🔹 463 tok/s for one user
🔹 223 tok/s/user at 8 concurrent users
Measured on multi-turn agentic coding workloads. Full 1M-token context supported.
📖 lightseek.org/blog/tokenspee…
⭐️ github.com/lightseekorg/t…
8× NVIDIA GB200 · NVFP4 + DFlash 2
🔹 463 tok/s for one user
🔹 223 tok/s/user at 8 concurrent users
Measured on multi-turn agentic coding workloads. Full 1M-token context supported.
📖 lightseek.org/blog/tokenspee…
⭐️ github.com/lightseekorg/t…
0
6
23
21,681
Frontier open intelligence meets next-generation NVIDIA compute.
Kimi K3 × NVIDIA Vera Rubin, with TokenSpeed MLA 2 accelerating long-context MLA decoding.
Thanks @KimiDevs for the support! 🚀
Kimi K3 × NVIDIA Vera Rubin, with TokenSpeed MLA 2 accelerating long-context MLA decoding.
Thanks @KimiDevs for the support! 🚀
Congrats to the TokenSpeed team! Building long-context agents with Kimi K3:
0
2
10
556
Introducing TokenSpeed MLA 2 for NVIDIA Vera Rubin 🚀
Kimi @Kimi_Moonshot K3 decode kernel speedups: up to 3.44× vs. MLA 1 on B200 and up to 2.84× on GB200, combining hardware + software gains.
Thanks to @NVIDIAAI for early access and engineering collaboration!
📖 lightseek.org/blog/tokenspee…
👉 github.com/lightseekorg/t…
Kimi @Kimi_Moonshot K3 decode kernel speedups: up to 3.44× vs. MLA 1 on B200 and up to 2.84× on GB200, combining hardware + software gains.
Thanks to @NVIDIAAI for early access and engineering collaboration!
📖 lightseek.org/blog/tokenspee…
👉 github.com/lightseekorg/t…
2
2
64
34,807
Day-0 support for Muse Spark Open on TokenSpeed.
190 output tokens/s per user on NVIDIA B300 — and that’s without speculative decoding.
Also supported on AMD Instinct MI355X.
👉 github.com/lightseekorg/t…
190 output tokens/s per user on NVIDIA B300 — and that’s without speculative decoding.
Also supported on AMD Instinct MI355X.
👉 github.com/lightseekorg/t…
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.
Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.
Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token ly multimodal across text, image, video, audio
2
5
17
44,727