Knowledge catalogue
Together AI has partnered with IBM and NVIDIA to launch the first dedicated, large‑scale AI inference cluster on IBM Cloud, consisting of NVIDIA B300 GPUs and Spectrum‑X Ethernet networking engineered
We just launched our own Jev-like classifier, together/Tev1-4B-experimental, on top of Qwen3.5 4B on Together’s serverless platform. In this blog post we’ll show you how to fine-tune your own version!
A global fintech uses Together’s Dedicated Model Inference (DMI) to run its GLM‑5.2 coding assistant, handling spike‑heavy traffic that peaks during engineering hours and cannot be met by static capac
Together Fine-Tuning adds the latest open-weight models, live experiment tracking, Expert LoRA, early stopping, tokenized dataset previews, pre-flight validation, and lower training prices on selected
In the DeepSWE benchmark a two‑stage approach of running DeepSeek V4 Pro 0813 first and falling back to GPT‑5.6 Sol when the tests fail achieves 83 % success at an average cost of 3.35 per task, compa
DeepSeek V4 Pro 0813 and Claude Fable 5 were benchmarked on DeepSWE. A cascading strategy that starts with Pro 0813 and escalates to Fable only when it fails solves 82.7 % of tasks at an average cost
DeepSeek‑V4 Flash 0731 is the cheapest model on the DeepSWE board, costing about 0.10 per rollout versus GPT‑5.6 Luna’s 0.61, yet it scores a pass@1 of 53.3% compared to Luna’s 67.2%. A cascade strate
**Kimi K3 is Moonshot AI’s 2.8‑trillion‑parameter open‑weight language model—the largest ever released—designed for frontier tasks such as long‑horizon coding and deep reasoning.** Its architecture us
GPU utilization can read healthy while your queue backs up, and a new replica takes minutes to warm. Here's how to pick autoscaling metrics, tune scale-up/down windows, and budget for cold starts on d
The Together AI platform’s dedicated inference architecture consists of three immutable entities: **configs** (engine, GPU type/count, parallelism and optimization profile), **deployments** (a specifi
ThunderAgent is a program-aware scheduler for agentic inference. By treating each agent workflow as a schedulable program, it eliminates KV cache thrashing to deliver more than 2x single-node throughp
OpenAI has updated its inference platform to give users full control over performance, cost, and quality without building their own stack—models go live in minutes and support multiple deployments beh
Together GPU Clusters now includes a suite of resilience and operational‑control features designed to address common large‑scale failure modes and team‑scaling challenges. The platform‑health improvem
Provisioned Throughput gives you reserved inference capacity for frontier open models like MiniMax M3 and GLM-5.2. Token-based pricing, a 99% uptime SLA, and up to 90% lower cost than proprietary APIs
Together AI announced an $800 million Series C funding round to support its mission of advancing open-source artificial intelligence development and deployment. The funding will be used to accelerate
Together AI presented research at ICML 2026 covering advances across the full technology stack, likely including foundational model improvements, inference optimization, and practical deployment solut
ParallelKernelBench is a benchmark from Together AI that evaluates frontier large language models' ability to write optimized multi-GPU CUDA kernels, revealing significant limitations in current LLMs'
Together AI has achieved ISO 27001:2022 certification, demonstrating its commitment to information security management and data protection standards required for enterprise AI deployments. This certif
MiniMax-M3 is a large language model capable of handling 1 million token contexts and multimodal inputs while maintaining efficient inference performance. Together AI's blog post discusses techniques
Together AI developed an optimized speech-to-text system focused on achieving the fastest processing speeds through technical innovations in their inference stack and model optimization. The approach
This article presents benchmarking results for AI coding agents evaluated at scale, likely comparing performance metrics such as code generation accuracy, execution success rates, and inference effici
Together AI and Pearl Research Labs announced a partnership aimed at lowering the cost of AI inference through collaborative research and development efforts. The partnership likely combines Together
Violin is an open-source video translation tool developed by Together AI that automatically translates video content to break down language barriers for viewers. The skill likely leverages AI models t
Together AI launched Voice Finder, a tool designed to help developers quickly select appropriate voices for their applications from a library of over 600 voice options. The tool streamlines the voice
Learn how to deploy any Hugging Face model in one session using Goose and Together's Dedicated Container Inference. Skip the setup complexity — one prompt gets your model running in a production-grade
DeepSeek-V4 makes million-token context a serving-systems problem. Together AI explores the inference work behind V4 on NVIDIA HGX B200, including compressed KV layouts, prefix caching, kernel maturit
This article from Together AI discusses foundational research techniques and methodologies used to enable efficient large-scale language model inference. It likely covers optimization strategies, hard
Together AI announced a strategic partnership with Adaption to advance AI capabilities and deployment. The collaboration likely focuses on integrating Adaption's technology or services with Together A
This article describes Together AI's experience shutting down a production service called 'Copy Fail,' detailing the operational and technical challenges of decommissioning a system that was only 732
DeepSeek-V4 Pro is now available on Together AI with 512K context, controllable reasoning modes, and cached-input pricing for long-context reasoning workloads like code agents, document intelligence,
Together AI announced immediate availability of NVIDIA's Nemotron 3 Nano Omni model to developers through its platform on the day of its release. The Nemotron 3 Nano Omni is a lightweight multimodal m
This article describes a technique for accelerating reinforcement learning (RL) rollouts using distribution-aware speculative decoding, which can achieve up to 50% speedup improvements. The method lik
This guide addresses the design and management of multi-tenant GPU clusters optimized for AI teams, focusing on strategies to maximize resource utilization while minimizing contention and conflicts be
Parcae is a stable looped language model that matches the quality of a Transformer twice its size — a 770M model reaching 1.3B-level performance. We introduce the first scaling laws for looping and sh
EinsteinArena is a platform where AI agents collaborate and compete on open math problems. AI agents on EinsteinArena have already set 11 new state-of-the-art results on open math problems — including