LiveOpenAI & Anthropic compatible · 98 models

Every model.
One endpoint.

Name: ModelStack API
Author: ModelStack

65 chat models and 33 image, video and audio models from 11 providers, behind one OpenAI-compatible API. Provider tiers take up to 80% off, and agent teams only pay for a specialist when a request needs one.

checkNo card required
checkOfficial rates or lower
checkSwitch models by changing a string

GET/v1/models200 OK

Claude Sonnet 5claude-sonnet-51M ctx$3 in · $15 outper 1M tokens
GPT-5.6gpt-5.61.05M ctx$4 in · $20 outper 1M tokens
Gemini 3.1 Pro Previewgemini-3.1-pro1.05M ctx$2 in · $12 outper 1M tokens
DeepSeek V3.2deepseek-v3.2160K ctx$0.26 in · $0.38 outper 1M tokens
Kimi K3kimi-k31M ctx$3 in · $15 outper 1M tokens
Grok 4.6grok-4.6500K ctx$2 in · $6 outper 1M tokens

+59 more chat models · 33 media modelsFull catalogarrow_forward

65
chat & reasoning models: 33
image, video & audio models: 11
providers behind one key: 80%
off at the economy tier

Drop-in compatible

Change one line. Keep the rest.

ModelStack speaks the OpenAI and Anthropic request formats, so your SDK, your streaming code and your tool-calling loop keep working. Point the base URL at us and pick any model by id.

swap_horizOpenAI and Anthropic request shapesUse either SDK as-is. Chat completions and Messages both work.
streamStreaming and tool calling pass throughAll chat completion parameters are forwarded — nothing to re-implement.
sync_altSwap models by changing a stringClaude to GPT to Gemini to DeepSeek: edit the model id, redeploy nothing.

main.py

1from openai import OpenAI
2 
3client = OpenAI(
4    api_key="sk_...",
5    base_url="https://api.modelstack.cc/v1",  # the only change
6)
7 
8response = client.chat.completions.create(
9    model="claude-sonnet-4-6",
10    messages=[{"role": "user", "content": "Hello"}],
11)

Provider tiers

Same models. Three prices.

Every request is routed through provider tiers, cheapest first, with automatic fallback when one is unavailable. The response is identical on every tier — only the rate changes.

Economy

0.20× provider cost

80% off

Lowest cost. Tried first on every request.

$10.00 of provider usage: $2.00
Pays with: Prepaid or bonus balance

Standard

0.50× provider cost

50% off

Balanced value. Next in line when economy is unavailable.

$10.00 of provider usage: $5.00
Pays with: Prepaid or bonus balance

Premium

1.00× provider cost

Official rate

Official provider APIs with the highest uptime. Final fallback.

$10.00 of provider usage: $10.00
Pays with: Prepaid balance only

Tiers are tried economy → standard → premium. Circuit breakers skip unhealthy providers so the request still succeeds.

How tiers and fallback workarrow_forward

One model id, a whole team behind it.

The coordinator answers routine requests and calls a specialist for the hard ones. Each specialist either advises the coordinator or replies to the user directly.

1Build it on a canvasaccount_treePick a coordinator model, add specialists from the roster, and set each one’s model, fallback and rules in place.
2Choose who writes the replycall_splitAdvise sends a specialist’s answer back to the coordinator to combine. Delegate sends it to the user exactly as written.
3Call it like a modelswap_horizUse the team id as the model. Same request format, same SDK, same streaming. Each agent bills at its own model’s rate.

account_treemodel: "my-coding-agent"AGENT TEAM

Coordinator answers68% of requests
1. Request
2. Coordinator
3. Reply
Senior Architect advises20% of requests
1. Request
2. Coordinator
3. Senior Architect
4. advisesCoordinator
5. Reply
Reviewer delegates12% of requests
1. Request
2. Coordinator
3. Reviewer
4. delegatesReply

Everything to the specialist: $5/M in
Blended with this split: ~$3.64/M in

Illustrative split for a coding workload. Rates are the models' list input prices.

Media generation

Image, video and audio on the same key.

Generation runs as async jobs: submit, poll, collect output URLs. Same API key, same balance — and available to your agents through MCP.

image

Image

12 models

Text-to-image, editing and upscaling.

Nano Banana 2
Nano Banana 2 Lite
Nano Banana Pro
GPT Image 2
GPT Image 1.5
+7 more

movie

Video

18 models

Text-to-video and image-to-video.

Veo 3
Veo 3 Fast
Veo 3 Lite
Kling 2.6
Kling 3.0
+13 more

graphic_eq

Audio

10 models

Text-to-speech, plus audio-driven video and lip-sync.

ElevenLabs Turbo 2.5
ElevenLabs Multilingual v2
Gemini 3.1 Flash TTS
Gemini 2.5 Pro TTS
Kling Avatar
+5 more

Model catalog

65 chat models from 11 providers.

Official rates or lower, on every tier. Assign any of them to a coordinator, a specialist, or a plain completion call.

Gemini
Google
12
GPT / o-series
OpenAI
11
Claude
Anthropic
9
Qwen
Alibaba
8
DeepSeek
DeepSeek
6
GLM
Zhipu
6
MiniMax
MiniMax
4
Kimi
Moonshot
4
MiMo
Xiaomi
2
Grok
xAI
2
Step
StepFun
1
All 98 modelsarrow_forward

Ecosystem

Works where you already work.

Any tool that talks to OpenAI or Anthropic talks to ModelStack. Every one below has a setup guide.

MCP server

Or hand your agent the whole toolbox.

One command connects Claude Code, Cursor or any MCP client. Your agent generates media and calls models on its own, billed to the key you already use.

auto_awesomeGenerate media mid-taskYour agent creates images, video and audio itself — no context switch, no separate dashboard.
savingsDelegate to cheaper modelsHand bulk work to a smaller model through one tool call and keep the expensive one for the hard parts.
searchDiscover models at runtimeLive catalogs of every model, the parameters each accepts and what it costs.

terminalclaude mcp add8 TOOLS

claude mcp add --transport http modelstack \
  https://api.modelstack.cc/mcp \
  --header "Authorization: Bearer $KEY"

generate_image
generate_video
generate_audio
get_generation
upload_media
llm_chat
list_models
list_generation_models

Same key, same balanceno new quota

Start with one line.

Free API key, no card. 98 models the moment you sign up.

Every model.One endpoint.