Together AI

Together AI releases new Qwen3-Next-80B models

Two new 80B parameter models are now available from Qwen on Together AI's platform. The models include specialized versions for thinking tasks and instruction following. This is an additive capability that expands model options without requiring any user action.

Date not specified

Together AI expands fine-tuning platform with 15 new large models and 131K token context

Together AI expanded fine-tuning support to include major new models from OpenAI, DeepSeek, Qwen, and Meta Llama families. Context length limits were significantly increased across existing models, with some reaching 131,072 tokens for SFT training. This enhancement provides developers with more powerful fine-tuning options and greater capacity for longer context training.

Date not specified

Together AI launches self-service GPU clusters with NVIDIA Blackwell and Hopper support

Together AI has released Instant Clusters allowing users to provision and manage GPU clusters through new API endpoints. The service supports NVIDIA's latest Blackwell (HGX B200) and Hopper (H100, H200) GPUs with scaling from single-node to hundreds of interconnected GPUs. This is a new feature addition that expands infrastructure capabilities but requires no immediate action from existing users.

Date not specified

Together AI expands evaluation with Serverless LoRA and Dedicated Endpoints

Together AI has expanded their evaluation capabilities to support Serverless LoRA models and Dedicated Endpoints, including fine-tuned models. This enhancement provides users with more flexibility in running evaluations across different model deployment types. No action required - this is an additive feature expansion.

Date not specified

Together AI releases Kimi-K2-Instruct-0905 – 1T Parameter MoE Model

Together AI has released an upgraded version of Moonshot's 1 trillion parameter Mixture of Experts model with enhanced performance capabilities. This is a new model addition that expands available AI capabilities without requiring any immediate action from users.

Date not specified

Together AI Deprecates 16 Models Including Llama & FLUX — Migrate to DeepSeek-V3.1

Together AI has deprecated 16 models across multiple providers including Meta Llama, Black Forest Labs FLUX, Qwen, and others, making them no longer available. Users must immediately migrate workflows using these deprecated models to alternative solutions. A new DeepSeek-V3.1 model with dual reasoning modes has been released as a potential replacement option.

Date not specified

Together AI adds fine-tuning support for GPT-OSS 20B and 120B models

Together AI now supports fine-tuning of OpenAI's open-source models (gpt-oss-20B and gpt-oss-120B) to create domain-specific variants. The feature includes support for 16K context supervised fine-tuning (SFT) and 8K context direct preference optimization (DPO). This is a new capability addition that enhances model customization options for users.

Date not specified

OpenAI GPT-OSS models now available on Together AI

Together AI has added OpenAI's first open-weight models (GPT-OSS-20B and GPT-OSS-120B) to their platform. This expands model availability for users but requires no immediate action. Users can now access these new open-weight options through existing Together AI integration.

Date not specified

Together AI launches VirtueGuard enterprise safety monitoring model

Together AI has launched VirtueGuard-Text-Lite, a new enterprise-grade model designed for real-time safety monitoring, content filtering, and bias detection. The model offers prompt injection protection with an 8ms response time for enterprise applications. This is an informational release of new functionality that expands Together AI's safety and monitoring capabilities.

Date not specified

Together AI launches LLM performance benchmarking framework

Together AI has released a new benchmarking platform that uses LLM-as-a-judge methodology to assess model performance. The framework supports custom evaluation suites with compare, classify, and score functionality for evaluating different models, prompts, and configurations. This is an informational release of new tooling that enhances model evaluation capabilities.

Date not specified

Together AI releases Qwen3-Coder-480B model with leading software engineering performance

Together AI has released a new large-scale coding model with 480B parameters and MoE architecture, featuring 256K context length for handling entire codebases. The model achieves top performance on SWE-Bench software engineering benchmarks and is designed for agentic coding applications. This is an additive capability that expands available model options without requiring any action from existing users.

Date not specified

Together AI adds NVIDIA HGX B200 support for DeepSeek-R1

Together AI now supports NVIDIA's Blackwell architecture (HGX B200) for the DeepSeek-R1 model, delivering significantly improved throughput and lower latency for serverless inference. The enhancement maintains existing API endpoints and pricing while providing record-breaking inference speeds. No action required from users as this is a performance improvement to existing infrastructure.

Date not specified

Together AI launches Kimi-K2-Instruct: 1T MoE model with agentic capabilities

Moonshot AI's new frontier-level model offers enhanced tool use, multi-step reasoning, and multilingual support through Together AI's platform. The model excels at function calling and agentic workflows, expanding available AI capabilities for developers. No action required - this is a new model addition to the existing service offerings.

Date not specified

Together AI launches Whisper Speech-to-Text APIs with enhanced performance

Together AI introduced new speech-to-text transcription APIs using OpenAI's Whisper Large v3 model, offering 15x faster processing than OpenAI's implementation with support for files over 1GB. The service includes speaker diarization, language detection, and multiple audio format support through standard OpenAI-compatible endpoints. This is a new capability addition that doesn't require any action from existing users.

Date not specified

Together AI achieves SOC 2 Type II compliance

Together AI has obtained SOC 2 Type II compliance certification through independent security audit, enabling easier vendor approval processes and support for regulated industries. This enhances the platform's enterprise readiness and reduces procurement friction for customers in regulated sectors. No immediate action required from existing users.

Date not specified