Fine-tune and serve open-weight models on your own cloud GPUs without managing ML infrastructure
Stage
Established
Not given
Founded
Not given
Not given
Based in
US
Not given
Pricing model
Not given
Freemium
Pricing
Not given
Free: $0 (1 training run per month, 1 model deployment, 1 member); Starter: $49/month (5 training runs per month, 5 model deployments, 1 member); Pro: $199/month (unlimited training runs, 1,000 model deployments, 10 included members, $30/month for each extra seat); Enterprise: custom pricing. Free i
Key features
GroqMetal: dedicated bare-metal infrastructure tuned for speed and reliability
GroqCore: production-ready inference stack with no infrastructure expertise required
GroqAssured: enterprise-grade governance, auditability, and control
256 LPUs per rack with 40 PB/s SRAM bandwidth
1,000 tokens per second per user throughput
128 GB on-chip SRAM per rack
315 PFLOPS of FP8 inference compute
13 globally distributed data centers across four continents
Train models with SFT, DPO, or RL using LoRA or QLoRA
Monitor training with real-time loss, time and cost tracking
Evaluate models by comparing base and tuned performance
Serve models as OpenAI-compatible API or wake-on-demand
Support for vision models including Gemma 4 and Qwen3.8
DPO preference learning from paired answers
Run budget-controlled experiments with hour and price limits
Download adapters as standard LoRA files or publish to Hugging Face
What makes it different
Fast inference at scale without sacrificing affordability
LPU pioneered technology integrated with NVIDIA GPUs
Fully integrated platform combining infrastructure, inference, and control
No training scripts required; workflow runs in console, CLI or via AI agent
GPU costs billed by your provider with zero markup from Tensorant
Data stays in your own storage; Tensorant keeps only workflow records
Free plan covers complete workflow on one model from import to API
Models are downloadable and portable to other platforms