ICEMIND
Disclaimer: This is a temporary placeholder site for ICEMIND. We are not selling any products at this time. The product is under active development; no offers are being made and no transactions will be processed on this domain.
Model Serving • Inference • Training | USA-based

Superhuman AI models. Cold precision.

ICEMIND delivers frontier-grade foundation models and an ultra-low-latency platform. Build, deploy and scale with enterprise-class security — without megacorp friction.

Launch Console Read Docs
< 12 ms
Token latency
> 99.95%
Availability
US-only
Data region
curl https://api.icemind.ai/v1/chat/completions \ -H "Authorization: Bearer $ICEMIND_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "ice-pro-2.1", "messages": [ {"role": "user", "content": "Plan a robotics workshop"} ], "stream": true }'
NASA
MIT
OpenGov
Axiom
Vertex
Quantica
Platform

A full-stack for intelligent products

From eval-driven training to ultra-fast inference, ICEMIND abstracts the complexity so teams ship in days, not months.

Turbo Inference

Token streaming under 12ms, CUDA graphs, paged KV-cache, tensor parallelism and speculative decoding by default.

Enterprise Guard

US-only data residency, SOC 2 Type II, private VPC peering, role-based controls, audit logs and content filters.

Custom Training

SFT/DPO/RLHF pipelines with synthetic data generation, eval gates and autoscaling GPU fleets (H100/B200).

Models

Frontier-grade model family

Pick the right tradeoff of reasoning, speed and context length. Drop-in via the ICEMIND API.

ICE-PRO-2.1

GA

Flagship reasoning model for complex tasks, tool-use and agents.

Context 256k • JSON mode • Function calling

API Docs → Benchmarks

ICE-LITE-1.8

GA

High-throughput model for chat, RAG and support at scale.

Context 128k • 2-4× cheaper • Low-latency

API Docs → Benchmarks

ICE-VISION-XL

GA

Multimodal model for image understanding, OCR and UI reasoning.

Vision+Text • Region prompts • OCR++

API Docs → Benchmarks
Benchmarks

State-of-the-art where it counts

Transparent, reproducible evals across public leaderboards and customer-defined tasks.

MMLU-Pro
84.1
GPQA-Diamond
62.7
Arena-Hard
Win 59%
MT-Bench
8.3

* Placeholder values — replace with your latest numbers & sources.

Developers

Build in minutes

Drop-in SDKs for TypeScript and Python.

// TypeScript import { ICEMind } from "icemind"; const icemind = new ICEMind({ apiKey: process.env.ICEMIND_KEY }); const stream = await icemind.chat.completions.create({ model: "ice-pro-2.1", messages: [{ role: "user", content: "Draft a product spec" }], stream: true, }); for await (const token of stream) process.stdout.write(token);
# Python from icemind import ICEMind cli = ICEMind(api_key=os.environ["ICEMIND_KEY"]) for token in cli.chat.completions.create( model="ice-pro-2.1", messages=[{"role":"user","content":"Summarize this contract"}], stream=True, ): print(token, end="")
Pricing

Simple, usage-based

Scale up or down without surprises. Custom enterprise plans available.

Builder

Monthly
$0.60 / 1M tokens
  • 100k free tokens
  • Community support
  • Shared GPUs
Get started →

Scale

Monthly
$1.80 / 1M tokens
  • SLA 99.95%
  • Private rate limits
  • Priority GPUs
Get started →

Enterprise

Annual
Custom
  • US-only data
  • SAML/SCIM
  • Dedicated clusters
Contact sales →

Ship with ICEMIND

Bring your roadmap. We bring the models, tooling and GPUs. Your users feel the speed immediately.

Talk to sales GitHub