Train in a Weekend Serve to Thousands
Compose experiments across SFT, Reinforcement Learning, and more on your own production code.
Use Synth with:
uvx synth-ai claudeFull-Stack Reinforcement Learning
Scalable pipelines for long-horizon GRPO.
Supervised Fine-Tuning
Spin up SFT jobs with curated datasets, auto-sharded compute, and full artifact tracking.
Supported Models
Qwen 3
Default for demos; supports tool-calling.
Qwen 3 (Advanced)
Enhanced variants with Instruct/Thinking modes and MoE support.
Qwen 3 Coder
Specialized for code generation.
Pricing
Only pay for the GPU time you burn.
It's Time to Train
Crafter SFT Loop
Collect traced rollouts, export JSONL, and launch a supervised job in minutes.
Qwen Coder LoRA
Run the 30B adapter playbook with Synth configs, compute guidance, and tuning tips.
Rejection Loop
Turn traced RL experience into curated JSONL, fine-tune, then evaluate the checkpoint.
Math RL
Deploy a math task app, run smoke tests, and stream a full on-policy training run.
Crafter On-Policy
Deploy a Crafter task app to Modal, verify health, and launch the production-style RL loop.
Evaluation Playbook
Run hosted evals, pull trace stats, and turn results into the next fine-tuning dataset.