Fine-tune and batch open models on GPUs without managing infrastructure
Stage
Established
Not given
Founded
Not given
Not given
Based in
US
Not given
Pricing model
Not given
Paid
Pricing
Not given
Fine-tuning from $0.15 per 1M training tokens (minimum $0.50 per job). Batch inference from $0.029 per 1M input tokens and $0.049 per 1M output tokens (minimum $0.20 per batch). Credit packs: Pack 1 ($10 for 10 credits), Pack 2 ($25 for 25 credits), Pack 3 ($100 for 100 credits). Custom amounts from
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
Fine-tune open models with LoRA, QLoRA or DPO methods
Run batch inference on base and fine-tuned models
Evaluate models against their base with custom graders
Download fine-tuned model adapters
Export merged weights of fine-tuned models
MCP server integration for Claude and other clients
OpenAI-compatible HTTP API
Real-time cost quotes before job execution
Per-token pricing with job minimums
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 GPU or infrastructure to manage; agents call the same API you would
Cost quotes before execution; no hidden charges beyond the quote
Download fine-tuned adapters; no hosting fees or endpoints to maintain
Works with Claude Code, Claude Desktop and any HTTP client without MCP dependency
Per-token pricing with no subscription or minimum contract