🤝Together AI Blog•Stalecollected in 18h
Together AI Adds Tool Calling, Reasoning, Vision to Fine-Tuning
💡Fine-tune tools, reasoning, VLMs—6x faster on 100B+ models now!
⚡ 30-Second TL;DR
What Changed
Native tool calling support
Why It Matters
This upgrade empowers developers to build advanced multimodal AI agents faster and cheaper. Higher throughput and estimates improve planning for production-scale deployments.
What To Do Next
Test fine-tuning a Llama-3.1 70B vision model on Together AI for 6x throughput gains.
Who should care:Developers & AI Engineers
Key Points
- •Native tool calling support
- •Reasoning model fine-tuning
- •Vision-language model support
- •100B+ parameter model training
- •Up to 6× higher throughput
🧠 Deep Insight
Background and context from public sources — not the original article. 8 sources cited.
🔑 Enhanced Key Takeaways
- •Together AI's fine-tuning platform now supports training models with 100B+ parameters including DeepSeek-V3 and Qwen3-235B, addressing a critical gap where competing platforms experience job failures at this scale[1][3]
- •The platform achieved up to 6× higher throughput through research-driven performance optimizations in ML systems, enabling faster iteration cycles and reducing time-to-production for fine-tuned models[1][3]
- •Integration with Hugging Face Hub allows users to directly read models from and write fine-tuned outputs to Hub repositories, reducing vendor lock-in and enabling seamless model sharing across the open-source ecosystem[1]
- •Together AI's reinforcement learning API exposes inference and training as separate, configurable layers with weight synchronization completing in under one minute at global distributed scale, directly addressing RL training bottlenecks[5]
📊 Competitor Analysis▸ Show
| Feature | Together AI | PremAI | WaveSpeedAI |
|---|---|---|---|
| Training data location | Together servers | Your cloud (S3/GCS/Azure) | Not specified |
| Fine-tuned model storage | Together infrastructure | Your cloud account | Not specified |
| Max model size supported | 100B+ (DeepSeek-V3, Qwen3-235B) | Not specified | Not specified |
| Tool calling support | Native | Not mentioned | Not mentioned |
| Vision-language models | Native support | Not mentioned | Specialized in visual generation |
| Inference throughput | Up to 6× optimization | Not quantified | Positioned as alternative |
| Vendor lock-in risk | High (data + models on platform) | Low (everything in your cloud) | Medium (partner-dependent for visual) |
| RLHF/DPO support | Full support with RL API | Not mentioned | Not mentioned |
🔮 Future ImplicationsAI analysis grounded in cited sources
Enterprise adoption of fine-tuned models will accelerate as Together AI's 100B+ parameter support and 6× throughput gains reduce training costs and iteration cycles from weeks to days
Existing customers report moving from weekly to daily iteration cycles with 2-3× cost savings, suggesting that removing infrastructure bottlenecks directly enables faster product development[3]
Open-source model ecosystems will consolidate around platforms with Hugging Face integration, as seamless Hub connectivity reduces friction for researchers and enterprises managing model versions
Together AI's Hub integration eliminates manual model export/import workflows, lowering barriers for teams to adopt and iterate on open-source models at scale[1]
Reinforcement learning workloads will shift toward cloud platforms as Together AI's RL API demonstrates that inference optimization research directly translates to 70%+ reduction in RL wall-clock time
By exposing rollout configuration and weight synchronization as tunable parameters, Together AI enables teams to optimize RL training loops rather than working around platform constraints[5]
⏳ Timeline
2025-09
Together AI expands fine-tuning platform with 100B+ model support, DPO options, and Hugging Face Hub integration
2026-01
Together AI announces reinforcement learning API with distribution-aware speculative decoding and ThunderAgent optimizations
2026-03
Together AI launches native support for tool calling, reasoning, and vision-language models in fine-tuning service with job cost and ETA estimates
2026-03
Together AI integrates NVIDIA Parakeet TDT 0.6B V3 ASR model for voice agent development
📎 Sources (8)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- together.ai — Fine Tuning Updates Sept 2025
- together.ai
- together.ai — Fine Tuning
- blog.premai.io — 19 Best Together AI Alternatives for Private Model Fine Tuning 2026
- together.ai — AI Native Conf Research and Product Announcements
- wavespeed.ai — Together AI Review 2026
- siliconflow.com — The Top Fine Tuning Platforms for Enterprises
- together.ai — Together AI at Nvidia Gtc 2026
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Together AI Blog ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.