The unlimited DeepSeek API grew into a fleet.
Stream still serves DeepSeek V4.1 Flash — newest public version, unlimited tokens, $5 a month per stream. Around it now stands a fleet of frontier models, and nothing in the fleet serves below the published intelligence floor. Same endpoint, same model ID, no meter.
What changed
One model became a fleet.
Stream launched as a flat-rate endpoint for a single model: DeepSeek V4 Flash. The deal was simple — unlimited tokens, one generation at a time, five dollars a month. That deal hasn't moved. What changed is what serves it: your stream is now paired with a model from a frontier fleet, DeepSeek among them, and every model has to clear the same public bar before it can serve you.
You still get DeepSeek
DeepSeek V4.1 Flash is in the fleet at its most recent publicly available version. The model ID you configure is simply auto.
You draw from a whole fleet
The current fleet is GPT-5.6 Luna, Muse Spark 1.3, GLM-5.3 Flash, and DeepSeek V4.1 Flash. It changes as new models launch.
Nothing serves below the floor
Every model in the fleet scores 70% or higher on Terminal-Bench 2.1, or 35 or higher on the Artificial Analysis Intelligence Index. Below the bar, it doesn't serve your stream.
Why flat rate
Stop engineering around the token bill.
Agent workloads read files, call tools, retry, and carry long histories. Their cost is difficult to predict because their work is difficult to predict.
No token meter
Use the fleet without a monthly token allowance or surprise overage line item.
Capacity you can understand
One generation per stream. Extra requests queue instead of increasing your bill. Stream quantities are fixed during the shutdown.
No GPU operations
We handle serving, routing, capacity, and recovery.
Common deployment patterns
Put the fleet to work.
Use Stream as your primary inference layer, an overflow path, or backup capacity.
Power your free tier
Offer useful AI features to every user without attaching an open-ended per-token cost to adoption.
Keep users going past limits
Route requests to Stream after premium-model credits run out, so users can keep working while you protect margins.
Run high-volume agents
Power request-heavy coding agents and autonomous tools like Hermes and OpenClaw without metering every loop.
Back up your main provider
Add a fallback route for outages or degraded service, keeping critical AI workflows available when your primary provider is not.
Coding agent compatibility
Bring your own agent.
Claude Code, Codex, Hermes, OpenCode, OpenClaw, Aider — if your coding agent accepts a custom OpenAI- or Anthropic-compatible endpoint, it can use Stream without changing the rest of your workflow.
Hermes
Keep tool calls, file operations, retries, and long agent sessions moving without metering every loop.
Read the setup guideOpenCode
Add Stream as an OpenAI-compatible provider and select auto from the model picker.
Read the setup guideOpenClaw
Use a custom model provider for always-on automations, coding workflows, and high-volume agent tasks.
Read the setup guideClaude Code
Point Claude Code at Stream's Anthropic-compatible gateway and map its model aliases to auto.
Read the setup guideCodex
Add a named Responses provider and launch Codex with a dedicated camelStream profile.
Read the setup guideAider
Export two environment variables and run Aider's terminal pair-programming loop through Stream.
Read the setup guideDrop-in by design
Keep your client. Change the endpoint.
The API follows the OpenAI format — streaming, tool calling, and structured output — plus Anthropic Messages for the agents that speak it.
from openai import OpenAI # Change the base URL and key.client = OpenAI( base_url="https://stream.camelai.com/v1", api_key="$CAMEL_API_KEY") response = client.chat.completions.create( model="auto", messages=messages, tools=tools, stream=True)Unlimited, said clearly
No token cap. A floor you can check.
Flat-rate inference only works when the boundaries are understandable. So we put them in the product instead of the fair-use fine print: capacity you can see, and a quality bar you can verify against public benchmarks.
Unlimited tokens
No monthly token allowance and no per-token overages.
One generation per stream
Additional requests queue within your existing stream capacity.
An intelligence floor
Every model that serves you scores 70% or higher on Terminal-Bench 2.1, or 35 or higher on the AA Intelligence Index.
Latest public versions
When your stream is paired with a model, you get its newest public release. Never a frozen snapshot.
260K context, guaranteed
Every request gets at least a 260K-token window — more when the model serving you supports it. Beyond it we compact the middle of the conversation; your task and latest turns stay intact.
24/7 access
Not a reserved daily time block. Generate whenever you need to.
Legacy plan · closed to new subscriptions
Existing customers keep their current capacity until their paid period ends. New subscriptions and stream increases are closed.
Existing customers: sign inYour subscription will not renew. Questions about your account? Contact us.
What's included
Frequently asked questions
About your existing account.
Check your remaining access.
Sign in to see your end date and invoices. Move connected agents to another provider before your access ends.
Existing customers: sign inQuestions first? Contact us.