Together AI
@togethercompute
Followers
49K
Following
2K
Media
517
Statuses
2K
AI pioneers train, fine-tune, and run frontier models on our GPU cloud platform.
San Francisco, CA
Joined November 2022
Introducing the Open Deep Research app!. Generate detailed reports on any topic with open source LLMs. Free & fully open source. We’re releasing everything: evaluation dataset, code, app, and blog.🔥
10
43
299
RT @Agentica_: It's easy to confuse Best@K vs Pass@K—and we've seen some misconceptions about our results. Our 59% on SWEBench-Verified….
0
15
0
RT @hardmaru: DeepSWE is a new state-of-the-art open-source software engineering model trained entirely using reinforcement learning, based….
0
42
0
Acknowledgements. DeepSWE stands on the shoulders of giants — it's trained from Qwen3-32B. Huge kudos to the.@Alibaba_Qwen team for open-sourcing such a powerful model!. This work is a joint collaboration between @Agentica_ and @togethercompute. Thanks to the @Agentica_ team and.
0
0
17
RT @Agentica_: 🚀 Introducing DeepSWE 🤖: our fully open-sourced, SOTA software engineering agent trained purely with RL on top of Qwen3-32B.….
0
64
0
Test-Time Scaling. We studied two approaches to boost agent performance at test time:. 📈 Scaling Context Length We expanded max context from 16K → 128K tokens. Performance improved, with ~2% gains beyond 32K, reaching 42.2% Pass@1. 🎯 Scaling Agent Rollouts We generated
1
0
19
Unlike reasoning models, RL for SWE agents creates unique challenges for scaling training environments. SWE-Agent Environments: Training with RL across parallel experiments spawns thousands of containers per RL iteration. We hit major bottlenecks with local Docker containers, so.
1
0
14
DeepSWE-Preview is a reasoning-enabled agentic model trained on the R2E-Gym training environments. Given a GitHub issue, the agent is given access to a repository as an RL environment and can use tools to explore the codebase, browse/view files, and use a bash terminal to
1
1
15
DeepSWE is trained directly with pure reinforcement learning—without distillation. In just 200 steps of training, its Pass@1 SWE-Bench-Verified score rises from 23→42.2%. Remarkably, RL improves DeepSWE’s generalization over time.
1
1
22
Links:. 📄 DeepSWE blog: 🤗 Model: 📄 ��rLLM blog:.💻 Github (rLLM):
1
6
34
Announcing DeepSWE 🤖: our fully open-sourced, SOTA software engineering agent trained purely with RL on top of Qwen3-32B. DeepSWE achieves 59% on SWEBench-Verified with test-time scaling (and 42.2% Pass@1), topping the SWEBench leaderboard for open-weight models. Built in
9
74
464
Together AI’s first GB200 cluster built by Dell!.
3
9
98
RT @vipulved: The first @togethercompute GB200 cluster CDUs imbibing coolant in prep to go live next week! Each rack here is 1.4 exaflops o….
0
11
0
Perfect for complex software engineering 🛠️, massive document analysis 🔍, and extended reasoning tasks 🧬. Deploy @MiniMax__AI's latest model via dedicated endpoints (40K/80K configs): .🔗
5
1
2
🧬⚡ MiniMax-M1 just landed on Together AI. The world's first open-weight hybrid-attention reasoning model is here:.🌌 1M token context.🧠 40K-80K token reasoning output.⚙️ 75% fewer FLOPs at scale.🎯 86.0% AIME 2024 performance. Your reasoning tasks just got more efficient 💎
7
10
54
Introducing our Together Demos page!. Explore our 24 free and open source AI apps to learn how to build your own. together.ai/demos
1
1
20
Our team will also be on the ground throughout the event. If you’re attending and would like to connect, book a meeting here 👇.
0
0
0
We’re excited to kick off @RaiseSummit next week in Paris!.Join our CEO, @vipulved, and @dylan522p, CEO of @SemiAnalysis_, for a conversation on Together AI’s vision for the future of AI infrastructure — happening July 8 at 9:00 AM.
1
2
10