zai-org · Language
GLM 5.3
zai-org/GLM-5.3
MoE · 780B params · 40B active
- Context
- 1M
- LoRA rank
- 16
- Fine-tune
- $4.60
- Serve
- $2.95
Nucleus helps you
build and own AI models
Private, custom AI models trained on your enterprise data, benchmarked against your standards, and hosted via API. Managed with dashboards, monitoring and automation. Frontier-level intelligence for your workflows, at a fraction of the operating cost.
Trusted by leading institutions
Product
Train. Evaluate. Serve. Own.
Build custom AI models on your data across six modalities, evaluate them against your benchmarks, serve them as private endpoints, and own every checkpoint. Three meters run the bill: training tokens, served tokens and storage.
49 open models, six modalitiesConnect your data
Connect your data
train on what matters
Bring your own JSONL or connect your data sources. Choose from open models across six modalities, with managed GPUs handling the rest.
- Connect your sources. SaaS platforms, databases, APIs, CSV files and document stores. Nucleus fetches your data and generates training-ready datasets.
- Automated training. Nucleus handles scheduling, GPU allocation, checkpointing and the production handoff. Set it up once and training runs on your schedule.
- Full loop access. Forward-backward passes and optimiser steps as first-class API calls. RL, DPO and distillation are yours to write.
Upload, build, or auto-generateKnow how good
Know how good
your model really is
Three ways to benchmark: upload your own test sets, build them in the dashboard, or let Nucleus generate them. Every checkpoint gets an intelligence index for your domain.
- Bring your benchmarks. Upload test sets that reflect your domain. Compare accuracy, speed and cost across checkpoints.
- Build in the dashboard. Define questions, criteria and pass thresholds for your workflows. Create evaluation sets without writing code.
- Auto-generated benchmarks. Nucleus generates domain-specific benchmarks, then validates each checkpoint with our LLM to score and verify the results.
Serverless, dedicated and reservedPrivate endpoints
Private endpoints
your SDK already speaks
Version your models, roll back to any checkpoint, and promote across environments. Swap your OpenAI or Anthropic base URL and the rest of your code stays the same.
- Serverless. Managed models at competitive rates. Idle time costs nothing.
- Dedicated. A GPU by the second, L40S to B200, with unlimited tokens on it. Cold starts are free.
- Reserved. One to six month terms at 15 to 25% below on-demand. Capacity guaranteed.
What you keep when you leaveEvery checkpoint
Every checkpoint
travels with you
Training data stays in your account. Weights are yours to export and run wherever policy requires.
- Weight export. Any checkpoint, any time, through the API. Run it on any hardware, in any region, behind any firewall.
- Data residency. Choose your region. Your cloud account by arrangement, or hardware you control.
- On-premise. Our GPU servers, located at your premises, maintained by us.
Models
Choose your base model
Open models across language, vision, image, video and audio. Fine-tune on your data, serve the result as a private endpoint, and keep the weights.
Qwen · Vision
Qwen3-VL 235B Instruct
Qwen/Qwen3-VL-235B-A22B-Instruct
MoE · 235B params · 22B active
- Context
- 128K
- LoRA rank
- 32
- Fine-tune
- $2.40
- Serve
- $1.60
black-forest-labs · Image generation
FLUX.2 dev
black-forest-labs/FLUX.2-dev
Flow transformer · 32B params
- Context
- Up to 4MP
- LoRA rank
- 32
- Fine-tune
- $1.40
- Serve
- $8.00
Lightricks · Video generation
LTX-2
Lightricks/LTX-2
DiT · native audio + video
- Context
- Up to 4K · 10s
- LoRA rank
- 32
- Fine-tune
- $1.60
- Serve
- $5.50
Qwen · Embedding & reranking
Qwen3 Embedding 8B
Qwen/Qwen3-Embedding-8B
Bi-encoder · 8B params · 4096 dims
- Context
- 32K
- LoRA rank
- 64
- Fine-tune
- $0.08
- Serve
- $0.05
openai · Language
GPT-OSS 120B
openai/gpt-oss-120b
MoE · 117B params · 5.1B active
- Context
- 128K
- LoRA rank
- 32
- Fine-tune
- $0.68
- Serve
- $0.45
Products
Own the whole journey
From dataset to deployed model: integrations that make your data trainable, managed GPUs, private endpoints and the APIs that tie them together.
Integrations
Connect your databases, APIs and document stores and turn them into training-ready datasets automatically.
- Databases, APIs and document stores
- Training-ready datasets
- Custom integration support
AI Cloud
Train on managed GPUs across six modalities, serve your models as private endpoints, and export the weights.
Custom AI Solutions
Fine-tuned models built on your enterprise data, evaluated and served as private endpoints.
- OpenAI + Anthropic compatible APIs
AI Chat Agents
Assistants for your teams, backed by your own custom-trained models and your private data.
APIs for custom solutions
Integrate your models into existing products with clean REST APIs, documented end to end.
Pricing
Near-frontier intelligence at 2% the cost
Three meters run the bill: training tokens, served tokens and storage. Reserve dedicated GPUs when you need guaranteed capacity. Idle time costs nothing.
- Training tokens
- from$0.03/ 1MFloor rate across 47 trainable bases.
- Served tokens
- from$0.02/ 1MInput and output tokens cost the same.
- Storage
- flat$0.10per GB / monthDatasets, checkpoints and exported weights.
- Dedicated GPUs
- from$1.99/ GPU-hourL40S to B200, billed by the second.
CompanyAlibabaNVIDIADeepSeekOpenAIZ AIGoogleKimiMistralMiniMaxMetaOtherreads more than one modalityunscored, shown in the lanewithdrawal scheduled
Use cases
Built with Nucleus
Custom models trained on Nucleus that connect to the tools a business already uses, and deliver the result a team used to spend a day on. Every workflow is production-ready and runs on your data.
47 trucks rerouted while they were still moving GPS traces, traffic and weather become live route changes, fuel efficiency per vehicle, and an SMS to every customer whose window slipped.
Quickstart
From dataset to a model you own
Five API calls take a JSONL file to a fine-tuned model. Two of them are the point: one serves it behind an OpenAI-compatible endpoint, the other hands you the weights. Prefer to drive the training loop yourself? Forward-backward passes and optimiser steps are first-class API calls too.
Click a step
Checkpoint000240
YoursYour applicationunchanged SDK
YoursYour hardwareruns anywhere
01_upload_dataset.sh
Bring a JSONL file of examples. That is the only prerequisite, and the file id it returns is what every later call refers to.
FilesTeams
Private models for every team
Fine-tune, evaluate, serve, and export models your organisation owns, with a workflow for every team that touches them.
ML Engineers
Fine-tune and deploy custom models on your enterprise data with full control over training pipelines and inference endpoints.
Research Teams
Drive the training loop call by call. RL, DPO, distillation, and custom losses, with the GPUs handled for you.
Product Teams
Ship features on models that speak your domain. Keep your existing OpenAI or Anthropic client code and swap the base URL.
Data Teams
Turn proprietary datasets into evaluated, versioned models with managed fine-tuning jobs. Upload a JSONL file, get back a private endpoint.
Platform Teams
Serve models behind OpenAI- and Anthropic-compatible endpoints, with usage tracking and rate limits built in.
Enterprise Leaders
Own your AI strategy. Training data stays in your account, weights are exportable, and models can run wherever policy requires.
Scale
Built to scale. Built to be owned.
Nucleus handles training and inference at any scale, from the first fine-tuning job to full production traffic, so your team ships models instead of managing GPUs.
- Your model weights
100% yours
- Compatible APIs
OpenAI + Anthropic
Tokens served
10.9M
+13.4%