Deep Research with open source models (+ open recipe)

Excited to share our work on scaling LLMs to handle million-token contexts! Training models for ultra-long sequences is challenging due to data scarcity. We introduce a novel hierarchical synthetic data generation pipeline to overcome this. Thrilled this will be presented at ICLR

Excited to share work from my Together AI internship—a deep dive into inference‑time scaling methods 🧠 We rigorously evaluated verifier‑free inference-time scaling methods across both reasoning and non‑reasoning LLMs. Some key findings: 🔑 Even with huge rollout budgets,

If you're at ICLR and passionate about optimizing language models for speed and efficiency, swing by the Together AI booth for a chat.

🔔 New blog post on how we can attain large speedups for our inference customers using custom speculators! 🚀 Key benefits of customization: ✅ ~1.3x faster inference ✅ ~25% cost reduction ✅ Gets better as you generate more responses

🚀 New research: YAQA — Yet Another Quantization Algorithm (yes, pronounced like yaca/jackfruit 🥭) Led by Albert Tseng, YAQA minimizes the KL divergence to the original model during quantization, cutting it by >30% vs. prior methods and outperforming even QAT on Gemma 3. 👇

.Together AI is building 2 gigawatts of AI factories (~100,000 GPUs) in the EU over the next 4 years with the first phase live in H2 '2025. AI compute is at <1% saturation relative to our 2035 forecast and we are starting early to build a large-scale sustainable AI cloud

Our open deep research app is launching in 24 hours! Generate reports about any topic using OSS LLMs. 100% free & open source.

How can we close the generation-verification gap when LLMs produce correct answers but fail to select them? 🧵 Introducing Weaver: a framework that combines multiple weak verifiers (reward models + LM judges) to achieve o3-mini-level accuracy with much cheaper non-reasoning

Open Deep Research app + fully open recipe ☺️

🔓⚡ FLUX.1 Kontext [dev] just landed on Together AI First open-weight model w/ proprietary-level image editing: 🎨 Perfect character consistency 🏆 Beats Gemini Flash + competitors 🛠️ Full model weights for customization Enterprise-quality editing, open weights💎

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

Come check out our poster on speeding up LLM, happening now til 1.30 pm. TL;DR — we show that we can hide the latency of all reduce operations in tensor parallel setting by modifying residual architecture to overlap MLP and attention.

TL;DR - one way to push the quality-efficiency frontier: obtain high quality generations via a collection of LLMs -> distill to a smaller model -> get a higher quality small model that is more inference-efficient than the original collection of models. Poster session

🤖OpenAI's open models are here. gpt-oss models just landed on Together AI. Achieves near-parity with o4- mini, trained using o3 techniques. Build anything, deploy anywhere🔥

LLM efficiency research on steroids with agentic workflows! 🚀

Most speculative decoding research focuses on algorithms. But we know that data matters a ton! (e.g. no matter how good the spec algorithm is, if it's trained on bad & misaligned data, the speed will be poor) What if we build on algorithms that make data really shine?! In