Together AI expands fine-tuning service with tool calling, reasoning, and vision support

2026-04-02 · Source: Together AI | The AI Native Cloud - Together.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Advanced, medium

Summary

Together AI has expanded its fine-tuning service, Together Fine-Tuning, with native support for tool calling, reasoning, and vision-language model (VLM) fine-tuning, effective March 18, 2026. This update addresses common reliability issues in multi-turn AI workflows, such as inconsistent tool calls and degraded reasoning. The service now supports training models with up to 1T parameters and datasets up to 100GB, delivering up to 6x higher throughput for larger models like Kimi-K2 and at least a 2x increase across all models. Enhancements include end-to-end fine-tuning for OpenAI-compatible tool schemas, specialized support for "thinking" tokens in reasoning traces, and native vision training for domain-specific visual data. Additionally, Together AI now provides job cost estimations before training and ETAs during training for better experiment planning.

Key takeaway

For AI Engineers building advanced multi-turn applications, Together AI's expanded fine-tuning capabilities directly address common reliability challenges. You can now fine-tune models for robust tool calling, complex reasoning, and domain-specific vision interpretation, reducing downstream failures. This allows your team to iterate faster, potentially cutting costs by 2-3x and improving accuracy from 77% to 87%, as demonstrated by XY.AI Labs. Leverage the new cost and time estimations to better plan your experiments and accelerate deployment.

Key insights

Together AI's fine-tuning service now natively supports complex AI workflows like tool calling, reasoning, and VLM training.

Principles

• Agentic AI requires reliable tool call execution.
• Structured reasoning traces improve model logic.
• Domain-specific visual data needs VLM alignment.

Method

The service enables fine-tuning on OpenAI-compatible tool schemas, "thinking" tokens for reasoning, and base64 encoded images for VLMs. It supports hybrid datasets and offers optional joint training of vision encoders and language layers.

In practice

• Fine-tune Qwen, Moonshot AI, Z.AI for tool calls.
• Train Qwen, Z.AI models on reasoning traces.
• Use hybrid datasets for VLM fine-tuning.

Topics

• LLM Fine-tuning
• Tool Calling
• Reasoning Traces
• Vision-Language Models
• AI Agent Development
• Model Training Optimization

Code references

• togethercomputer/together-cookbook
• Dao-AILab/sonic-moe

Best for: AI Architect, Computer Vision Engineer, CTO, Machine Learning Engineer, AI Engineer, MLOps Engineer

Related on AIssential

• Implementing CoT Fine-Tuning for Lending — CoT fine-tuning transforms black-box LLMs into auditable reasoning tools for high-stakes financial decisions.
• Failure Makes the Agent Stronger: Enhancing Accuracy through Structured Reflection for Reliable Tool Interactions — Explicitly training LLMs to reflect on and correct tool-use errors significantly improves multi-turn interaction reliability.
• Introducing Mistral Small 4 — Mistral Small 4 unifies multiple specialized AI capabilities into a single, versatile model.
• Paper Digest: ICLR 2026 Papers & Highlights — The ICLR 2026 highlights reveal a strong focus on enhancing LLM reasoning, multi-modal capabilities, and robust AI agent development.
• WHAT? Qwen3.6-A3B Could Solve This? — Instruction tuning and reasoning trace analysis can significantly enhance small MoE model performance on complex tasks.
• The End of Monolithic AI: BIG + Small LM Together — Duet uses two LLMs, one for reasoning and one for generation, to reduce inference costs by up to 70%.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Together AI | The AI Native Cloud - Together.ai.