Model specifications and this workspace
The official model card describes a 1M-token context window, 1.05T total parameters and 52B active parameters. Text and images are supported. Those model specifications do not enlarge this workspace’s request limits. Email/password accounts work here; real model generation, Google sign-in, billing and cloud sync remain pending configuration.
Your first conversation
Choose a sample prompt or write a focused question in the playground. Add the source material and adjust the settings. Continue to chat with the same draft and settings. Sign in using an email and password when asked; email verification is not required. In a local demo, the response is a labelled fixed example. On the live site, generation remains unavailable until the provider is connected.
Write a useful prompt
Specify the goal, source and output. For documents, number passages and ask for quoted evidence. For code, include the failing check, expected behavior and relevant functions. For writing, name the audience, language, tone and terminology. Extract PDF text before pasting it. Ask the model to mark missing evidence instead of filling gaps.
Goal: explain this function to a new teammate.
Source: [paste the function and its caller]
Output: purpose, edge cases, one minimal fix, and checks.
If context is missing, state exactly what you need.Responses, drafts and history
Enter sends or continues the draft; Shift + Enter adds a line. Image URLs and model settings travel with a playground draft into chat. Completed chat conversations are stored in this browser and can be copied or exported. Refreshing chat preserves the browser history. Cloud sync is unavailable on the live site. Keep an export before clearing browser storage.
Call the provider API
The examples on this site use Mistral’s provider API, not an account API key issued by this workspace. Configure MISTRAL_API_KEY on your server. Use https://api.mistral.ai/v1 with an OpenAI-compatible client, or the Mistral SDK. The published model alias is mistral-large-4. Examples are integration references and have not been executed against a paid provider in this build.
import os
from mistralai.client import Mistral
client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])
response = client.chat.complete(
model="mistral-large-4",
messages=[{"role": "user", "content": "Explain retries."}],
max_tokens=1024,
)
print(response.choices[0].message.content)Request parameters and limits
The workspace accepts 1–24 user/assistant messages with 1–24,000 characters each. A system prompt accepts up to 4,000 characters. Output is 64–4,096 tokens; temperature is 0–2; reasoning is none or high; output is text or JSON. These are application limits. Provider API limits depend on the endpoint and model. Higher reasoning can consume more of the output budget.
Read streaming responses
For the provider API, enable stream and handle both plain text and structured content chunks. Buffer raw SSE bytes until a complete event arrives. Stop closes the local request; failed and stopped generations release the workspace credit reservation. Treat partial JSON and tool arguments as incomplete. The workspace endpoint uses its own delta/done/error events; it is not an OpenAI-compatible public API.
Images and structured output
Attach a public HTTPS image URL without credentials. This workspace does not upload PDFs, audio or video. Describe the question beside the image and check small labels against the original. JSON mode needs a prompt that asks for JSON; parse and validate the response before using it. A valid JSON object alone does not establish that its claims are correct.
Tool calling and agents
Define functions in the provider API integration. The model proposes names and JSON arguments. Your server validates schema, access and side effects before executing a function, then returns a tool result using the matching tool_call_id. A chat prompt here does not run a function or give an agent access to your files.
# Configure your server environment
export MISTRAL_API_KEY="your-provider-key"
export MISTRAL_MODEL="mistral-large-4"
# For an OpenAI-compatible client
export OPENAI_BASE_URL="https://api.mistral.ai/v1"
export OPENAI_API_KEY="$MISTRAL_API_KEY"
# Keep execution and permissions in your application.Credits, plans and cost
Proposed Free allowance is 100 credits per day, resetting at midnight UTC+8. Proposed Pro is $19/month with 12,000 credits per billing period; grants replace unused paid credits. One credit corresponds to $0.0001 of configured model cost, with completed usage rounded up. Failed or stopped generations are not charged. Checkout is unavailable. Provider API costs are separate; use the calculator for that budget.
Weights and deployment
The release announcement schedules weights for the end of October 2026. Confirm the actual repository, license and serving support when the files arrive. Total parameter storage, KV cache and runtime overhead all matter. Active parameter count is not a weight-memory estimate. Read the deployment guide before choosing hardware.
Troubleshooting and support
401: sign in again and preserve the draft. 400: check message length, image URL and settings. 429: wait before retrying. 503: the service is not configured or temporarily unavailable; an account alone does not enable model generation. An interrupted answer is not complete. Record the page, time and error; omit passwords, keys and private documents when contacting support.