The Curious Engineer

All blogs / AI / ML

Together.ai

together.ai/blog

Explore Together AI's latest product launches, research papers, and technical guides on AI inference, fine-tuning, and GPU infrastructure.

AI / ML

’22’23’24’25’26
Posts
259 from 17 Nov 2022 to 27 Oct 2026
Posting
weekly (98 in the last year)
Kind
company blog
Feed
https://www.together.ai/blog/rss.xml

2026 86

  1. Auto repair for Together GPU Clusters: from GPU fault to healthy node
  2. Together Post-Training: frontier-grade model shaping, the way that fits you
  3. Meta Muse Spark Open is live on Together AI
  4. Expanding our enterprise inference capacity with IBM Cloud and NVIDIA
  5. Together Link: open models in the harness you already use. Start with one command today.
  6. How to train your own Jev for $17
  7. Canary rollouts: upgrade models in production without downtime
  8. How a global fintech scaled coding agent traffic with Dedicated Model Inference
  9. Migrating from closed to open source models, Together
  10. Together AI expands fine-tuning service with more models, live metrics, and finer controls
  11. Introducing preemptible compute: the same compute, half the price
  12. To Infinity and Beyond: ThunderKittens Now on NVIDIA Vera Rubin NVL72!
  13. The Open Source AI Stack
  14. GLM-5.3 vs. GLM 5.3 Flash on DeepSWE →
  15. GLM-5.3 vs. GPT-5.6 Sol on DeepSWE: Cost, Coding, and Routing
  16. GLM-5.3 vs. Claude Fable 5 on DeepSWE: Cost, Coding, and Routing
  17. Model LibraryDeepSeek V4 Pro 0813 vs GPT-5.6 Sol on DeepSWE: Cost, Coding, and RoutingWe ran 904 DeepSWE rollouts on DeepSeek V4 Pro 0813 and GPT-5.6 Sol. Sol leads pass@1 by 10 points at 35x the cost; Pro wins pass@4, and a Pro-first cascade hits 83.0%.
  18. DeepSeek V4 Pro 0813 vs Claude Fable 5 on DeepSWE: Cost, Coding, and Routing
  19. A/B test models in production
  20. Model LibraryDeepSeek-V4 Flash 0731 vs GPT-5.6 Luna on DeepSWE: Cost and CodingWe ran 900 DeepSWE rollouts on DeepSeek-V4 Flash and GPT-5.6 Luna. Luna leads pass@1 by 14 points; DeepSeek delivers 4.8x the solves per dollar.
  21. Kimi K3: the complete developer guide
  22. InferenceAutoscaling endpoints for LLM inferenceGPU 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 dedicated inference.
  23. Together AI announces strategic partnership with Moonshot AI to natively serve Kimi models
  24. InferenceConfiguring Dedicated Model InferenceThe three-part resource model behind Together AI Dedicated Model Inference—endpoints, deployments, configs—and how capacity-aware routing ties them together.
  25. ThunderAgent: 2x Faster Agentic Inference for Synthetic Data Generation at Scale
  26. Model LibraryKimi K3 vs GPT-5.6 Sol on DeepSWE: Cost, Coding, and RoutingWe ran 904 DeepSWE rollouts on Kimi K3 and GPT-5.6 Sol. Sol leads pass@1; Kimi K3 wins pass@4 at 2.8x the solves per dollar, and routing between them reaches ~85.6%.
  27. Model LibraryKimi K3 vs Claude Fable 5 on DeepSWE: Cost and CodingWe ran 452 DeepSWE rollouts on Kimi K3 and Claude Fable 5. Fable leads pass@1 by 1.4 points; Kimi K3 wins pass@4 and delivers 2.8x the solves per dollar.
  28. The production platform for open-weight AI inference
  29. Together AI and Y Combinator partner to launch the first dedicated GPU cluster for the YC community
  30. What does 99.9% uptime mean for inference?
  31. Together AI brings Thinking Machines Lab’s new model Inkling on day 0
  32. New in Together GPU Clusters: Reliability and control for production GPU clusters
  33. InferenceOpen, convenient and predictable: Introducing Provisioned ThroughputProvisioned 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. No GPU-hour math, no
  34. Announcing our $800M Series C to accelerate the shift to open-source AI
  35. Together AI at ICML 2026: frontier research across the full stack
  36. ParallelKernelBench: Frontier LLMs can't write fast multi-GPU kernels (yet)
  37. Kimi K2.7 Code vs Claude Fable 5: Landing pages that cost 94% less
  38. Building trust in enterprise AI: Together AI earns ISO 27001:2022 certification
  39. Serving MiniMax-M3 for efficient inference: Unlocking 1M-Token Context and Multimodality Without Regrets
  40. How Together AI built the world’s fastest speech-to-text stack
  41. Benchmarking inference at scale: coding agents
  42. Together AI and Pearl Research Labs Team Up to Reduce the Cost of AI Inference
  43. Violin: An open-source video translation skill that breaks language barriers
  44. Introducing voice finder a new tool to quickly find the right voice for your app from over 600 voices
  45. Serving DeepSeek-V4: why million-token context is an inference systems problem
  46. Deploy and inference any model from HuggingFace
  47. Foundational research powering efficient inference at scale
  48. From 732 bytes to nowhere: shutting down Copy Fail in production
  49. Announcing Together AI and Adaption Partnership
  50. DeepSeek-V4 Pro now available on Together AI
  51. Together AI Brings NVIDIA Nemotron 3 Nano Omni to Developers on Day 0
  52. Accelerate RL rollouts by up to 50% with distribution-aware speculative decoding
  53. Capacity without conflict: A guide to multi-tenant GPU cluster design for AI-native teams
  54. Parcae: Doing more with fewer parameters using stable looped models
  55. EinsteinArena: Harnessing the collective intelligence of agents in the wild to advance science
  56. What is an AI Native Cloud?
  57. Wan 2 7 now available on together ai
  58. AI for Systems: Using LLMs to Optimize Database Query Execution
  59. Deepgram speech to text and voice models now available natively on together ai
  60. Inside the Together AI kernels team
  61. Aurora
  62. Plan, divide, and conquer: How weak models excel at long context tasks
  63. Together AI expands fine-tuning service with tool calling, reasoning, and vision support
  64. Mamba-3
  65. Together AI at NVIDIA GTC 2026: Explore our latest innovations across research and products
  66. Build real-time voice agents on Together AI
  67. Together AI Brings NVIDIA Nemotron 3 to Developers on Day 0
  68. New in Together GPU Clusters: Autoscaling, observability, and self-healing
  69. Key research and product announcements at the AI Native Conf
  70. FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware Scaling
  71. Cache-aware prefill–decode disaggregation (CPD) for up to 40% faster long-context LLM serving
  72. Introducing Together AI’s new look
  73. Coderforge preview
  74. How speech models fail where it matters the most and what to do about it
  75. Consistency diffusion language models: Up to 14x faster inference without sacrificing quality
  76. Introducing Dedicated Container Inference: Delivering 2.6x faster inference for custom AI models
  77. What do LLMs think when you don't tell them what to think about?
  78. Rime arcana v3 turbo and rime arcana v3 now available on together ai
  79. Together AI welcomes Alon Gavrielov as VP of Infrastructure Strategy
  80. Fine tuning open llm judges to outperform gpt 5 2
  81. Together Evaluations now supports comparing top commercial APIs vs. open source models
  82. DSGym: A holistic framework for evaluating and training data science agents
  83. Optimizing inference speed and costs: Lessons learned from large-scale deployments
  84. Learn how Cursor partnered with Together AI to deliver real-time, low-latency inference at scale
  85. Inside multi-node training: How to scale model training across GPU clusters
  86. How to choose the right open model for production

2025 63

  1. Minimax speech 2 6
  2. Rime voice models now available on together ai
  3. Research POV: Yes, AGI Can Happen – A Computational Perspective
  4. Announcing native availability of NVIDIA Nemotron 3 Nano, NVIDIA’s latest reasoning model
  5. Announcing Together Python SDK v2.0
  6. Together AI and Meta partner to bring PyTorch Reinforcement Learning to the AI Native Cloud
  7. How to run TorchForge reinforcement learning pipelines in the Together AI Native Cloud
  8. Introducing AutoJudge: Streamlined inference acceleration via automated dataset curation
  9. Together AI delivers fastest inference for the top open-source models
  10. Flux 2 multi reference image generation now available on together ai
  11. The fastest inference for realtime voice ai agents
  12. How to evaluate and benchmark Large Language Models (LLMs)
  13. Dynamic AI agent testing for the real world with Collinear Simulations and Together Evals
  14. Large Reasoning Models Fail to Follow Instructions During Reasoning: A Benchmark Study
  15. 40 new image and video models
  16. Announcing the Together AI Startup Accelerator, purpose-built for AI Native Apps
  17. AdapTive-LeArning Speculator System (ATLAS): A New Paradigm in LLM Inference via Runtime-Learning Accelerators
  18. Improved Batch Inference API: Enhanced UI, Expanded Model Support, and 3000× Rate Limit Increase
  19. Fine-Tuning Platform Upgrades: Larger Models, Longer Contexts, Enhanced Hugging Face Integrations
  20. Together AI welcomes Mahadev Konar as SVP for Infrastructure Engineering
  21. Deepseek v3 1 hybrid thinking model now available on together ai
  22. How Together AI Uses AI Agents to Automate Complex Engineering Tasks: Lessons from Developing Efficient LLM Inference Systems
  23. Transform OpenAI gpt-oss Models into Domain Experts with Together AI Fine-Tuning
  24. Fine-Tuning Small Open-Source LLMs to Outperform Large Closed-Source Models by 60% on Specialized Tasks
  25. OpenAI's New Open gpt-oss Models vs o4-mini: A Real-World Comparison
  26. Announcing the Availability of OpenAI's Open Models on Together AI
  27. Together Evaluations: Benchmark Models for Your Tasks
  28. Qwen3-Coder: The Most Capable Agentic Coding Model Now Available on Together AI
  29. Together AI Delivers Top Speeds for DeepSeek-R1-0528 Inference on NVIDIA Blackwell
  30. Back to The Future: Evaluating AI Agents on Predicting Future Events
  31. Kimi K2: Leading Open-Source Model Now Available on Together AI
  32. Together AI Launches Speech-to-Text: High-Performance Whisper APIs
  33. Powering Secure AI: Together AI Achieves SOC 2 Type 2 Compliance
  34. DeepSWE: Training a Fully Open-sourced, State-of-the-Art Coding Agent by Scaling RL
  35. From Zero to One: Building An Autonomous and Open Data Scientist Agent from Scratch
  36. Introducing the Together AI Batch API: Process Thousands of LLM Requests at 50% Lower Cost
  37. The Frontier is Open
  38. FLUX.1 Kontext models: Character consistency and precise image editing without fine-tuning
  39. Mixture-of-Agents Alignment: Harnessing the Collective Intelligence of Open-Source LLMs to Improve Post-Training
  40. Together Code Interpreter: execute LLM-generated code seamlessly with a simple API call
  41. Together Code Sandbox: the most robust infrastructure for building AI coding products at scale
  42. Introducing Together Code Sandbox & Together Code Interpreter: SOTA code execution for AI
  43. Boosting DeepSeek-R1’s Speed with Customized Speculative Decoding
  44. Salesforce, Zoom, InVideo Train Faster with Together AI Turbocharged with NVIDIA Blackwell
  45. Chipmunk: Training-Free Acceleration of Diffusion Transformers with Dynamic Column-Sparse Deltas
  46. Continued Fine-tuning of LLMs: A Technical Deep Dive
  47. Direct Preference Optimization: A Technical Deep Dive
  48. Together Fine-Tuning Platform, Now With Preference Optimization and Continued Training
  49. Open Deep Research
  50. DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level
  51. Together AI partners with Meta to offer Llama 4: SOTA Multimodal MoE Models
  52. Together AI Awarded ClusterMAX™ Gold Rating by SemiAnalysis
  53. Introducing Together Instant GPU Clusters Accelerated by NVIDIA GPUs, with Self-Service Provisioning in Minutes
  54. Deploy Leading AI Models Accelerated by NVIDIA NIM on Together AI
  55. On-demand dedicated endpoints: run inference with unmatched price-performance & control at scale
  56. Together AI becomes NVIDIA Cloud Partner to bolster accelerated AI offerings
  57. Minions: embracing small LMs, shifting compute on-device, and cutting cloud costs in the process
  58. Together AI Achieves 90% Faster BF16 Training with NVIDIA Blackwell Platform and Together Kernel Collection
  59. Deploy DeepSeek-R1 at scale: Fast, secure serverless APIs and large-scale Together Reasoning Clusters
  60. How to deploy DeepSeek-R1 and distilled models securely on Together AI
  61. Mistral Small 3 API now available on Together AI: A new category leader in small models
  62. Generate images with specific styles using Flux LoRAs on Together AI
  63. Build ultra low latency voice AI applications with Together AI and Cartesia Sonic

2024 44

  1. Together AI acquires CodeSandbox to launch first-of-its-kind code interpreter for generative AI
  2. Announcing Llama 3.3 70B, with enhanced reasoning, mathematics, and instruction-following on Together AI
  3. Fine-tuning API: Introducing long-context training, conversation data support and more configuration options
  4. Fine-Tuning LLMs for Multi-Turn Conversations: A Technical Deep Dive
  5. Long Context Fine-Tuning: A Technical Deep Dive
  6. FLUX Tools now available via Together APIs: Get greater control over image generation with Canny, Depth and Redux models
  7. Together AI to Co-Build Turbocharged NVIDIA GB200 Cluster with 36K Blackwell GPUs in Partnership with Hypertec Cloud
  8. Even Better, Even Faster Quantized LLMs with QTIP
  9. Linearizing LLMs with LoLCATs
  10. How to build a real-time image generator with Flux and Together AI
  11. Multimodal Document RAG with Llama 3.2 Vision and ColQwen2
  12. FLUX API is now available on Together AI: New FLUX1.1 [pro] and free access to FLUX.1 [schnell]
  13. Together AI launches Llama 3.2 APIs for vision, lightweight models & Llama Stack: powering rapid development of multimodal agentic apps
  14. Introducing The Together Enterprise Platform: Run GenAI securely in any environment, with 2x faster inference and continuous model optimization
  15. The Mamba in the Llama: Distilling and Accelerating Hybrid Models
  16. Supercharging NVIDIA H200 and H100 GPU Cluster Performance With Together Kernel Collection
  17. Speculative decoding for high-throughput long-context inference
  18. TEAL: Training-Free Activation Sparsity in Large Language Models
  19. A practitioner's guide to testing and running large GPU clusters for training generative AI models
  20. Llama 3.1: Same model, different results. The impact of a percentage point.
  21. Together AI partners with Meta to release Llama 3.1 models for inference and fine-tuning with accelerated performance at full accuracy
  22. Together AI and NVIDIA collaborate to power Llama 3.1 models for enterprises on NVIDIA DGX Cloud
  23. Fine-tuning Llama-3 to get 90% of GPT-4’s performance at a fraction of the cost
  24. FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision
  25. Building a personalized code assistant with open-source LLMs using RAG Fine-tuning
  26. SpecExec: Massively Parallel Speculative Decoding for Interactive LLM Inference on Consumer Devices
  27. Dragonfly: A large vision-language model with multi-resolution zoom
  28. Using Axiomic to build multi agent chat with Together API
  29. Flash Attention received the inaugural Stanford open source software award
  30. ThunderKittens: A Simple Embedded DSL for AI kernels
  31. FAQ: Building LLMs with RedPajama-v2, a 30 trillion token web dataset
  32. Together AI partners with Snowflake to bring Arctic LLM to Enterprise customers
  33. Announcing v1 of our Python SDK
  34. Announcing $106M round led by Salesforce Ventures
  35. Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding
  36. BASED: Simple linear attention language models balance the recall-throughput tradeoff
  37. Evo: Long-context modeling from molecular to genome scale
  38. BitDelta: Your Fine-Tune May Only Be Worth One Bit
  39. Announcing function calling and JSON mode
  40. Introducing the Together Embeddings endpoint — Higher accuracy, longer context, and lower cost
  41. Long context retrieval models with Monarch Mixer
  42. Building your own RAG application using Together AI and Langchain
  43. Building your own RAG application using Together AI and LlamaIndex
  44. Building your own RAG application using Together AI and MongoDB Atlas

2023 32

  1. Mamba-3B-SlimPJ: State-space models rivaling the best Transformer architecture
  2. Can you feel the MoE? Mixtral available with over 100 tokens per second through Together Platform!
  3. Filter responses of any model with Llama Guard or your own safety model
  4. Paving the way to efficient architectures: StripedHyena-7B, open source models offering a glimpse into a world beyond Transformers
  5. Our $102.5M Series A
  6. Growing to 20 exaflops, Together GPU Clusters help startups and enterprises accelerate generative AI development
  7. Announcing Together Custom Models. Build a state-of-the-art LLM with Together AI — and own the model.
  8. FlashFFTConv: Efficient Convolutions for Long Sequences with Tensor Cores
  9. RedPajama-Data-v2: An open dataset with 30 trillion tokens for training large language models
  10. Flash-Decoding for long-context inference
  11. Medusa: Simple framework for accelerating LLM generation with multiple decoding heads
  12. Llama-2-7B-32K-Instruct — and fine-tuning for Llama-2 models with Together API
  13. Faster inference enables up to 5x price reduction on Together API
  14. Preparing for the era of 32K context: Early learnings and explorations
  15. Monarch Mixer: A new model architecture for increased efficiency
  16. Together AI and Snorkel AI empower enterprises to build proprietary LLMs
  17. Together AI launches full stack for developers to build with open-source AI
  18. RedPajama 7B now available, instruct model outperforms all open 7B models on HELM benchmarks
  19. Decentralized training of foundation models in heterogeneous environments
  20. Fine-tuning language models over slow networks using activation compression with guarantees
  21. FlashAttention: Fast and memory-efficient exact attention with IO-Awareness
  22. Together’s $20M seed funding to build open-source AI and cloud platform
  23. RedPajama-INCITE-3B, an LLM for everyone
  24. Releasing 3B and 7B RedPajama-INCITE family of models including base, instruction-tuned & chat models
  25. CocktailSGD: Fine-tuning foundation models over 500Mbps networks
  26. RedPajama training progress at 440 billion tokens
  27. RedPajama, a project to create leading open-source models, starts by reproducing LLaMA training dataset of over 1.2 trillion tokens
  28. OpenChatKit now runs on consumer GPUs with a new 7B parameter model
  29. FlexGen: High-throughput generative inference of large language models with a single GPU
  30. Announcing OpenChatKit
  31. Hyena Hierarchy: Towards larger convolutional language models
  32. FlashConv: Speeding up state space models

2022 5

  1. Hungry Hungry Hippos: Towards language modeling with state space models
  2. NeurIPS 2022: Overcoming communication bottlenecks for decentralized training (2/2)
  3. NeurIPS 2022: Overcoming communication bottlenecks for decentralized training (1/2)
  4. Releasing GPT-JT powered by open-source AI
  5. HELM: benchmarking large language models on the Together Research Computer

Undated 29

  1. Best practices to accelerate inference for large-scale production workloads
  2. From AWS to Together Dedicated Endpoints: Arcee AI's journey to greater inference flexibility
  3. Scaling AI Companions: How Dippy AI Reached Over 4 Million Tokens/Minute with Together Dedicated Endpoints
  4. How to fine-tune an LLM: a developer's guide
  5. How Zomato built an AI customer support bot that doubled customer satisfaction and scaled to over 1,000 messages per minute
  6. KV cache explained: memory math, GQA, and PagedAttention
  7. LLM inference: How it works and how to optimize it
  8. Optimizing Training Workloads for GPU Clusters
  9. RAG vs. fine-tuning: a decision framework for LLM apps
  10. How to Build a State-of-the-Art Search Stack for LLMs: RAG, Reranking, and Reinforcement Learning
  11. Thunderkittens nvidia blackwell gpus
  12. Together ai acquires refuel ai
  13. Together ai announcing 305m series b
  14. Together ai available aws marketplace to accelerate enterprise ai development
  15. Together ai expands in europe
  16. Together ai partners with meta to release meta llama 3 for inference and fine tuning
  17. Together ai powers pioneers at nvidia gtc 2025
  18. Together ai welcomes kai mak
  19. Together chat
  20. Together crusoe reduce carbon impact of generative ai
  21. Together inference engine 2
  22. Together inference engine v1
  23. Together instant clusters ga
  24. Together moa
  25. Together rerank api and salesforce llamarank
  26. Tri dao flash attention
  27. Virtueguard
  28. What are llms
  29. Yaqa