DeepSeek V3.2 is DeepSeek's efficient reasoning model, released 2025-12, built to balance strong reasoning and agentic tool use with low computational cost. It introduces DeepSeek Sparse Attention, a fine-grained sparse attention mechanism that cuts training and inference cost while preserving quality on long contexts. A scalable reinforcement-learning post-training framework lifts reasoning into the GPT-5 class, and the model earned gold-medal results at the 2025 IMO and IOI. A large-scale agentic task synthesis pipeline further improves tool-use compliance and generalization.
This model is not routed on MirAPI yet — prices shown are the official list prices. Create an account to get notified when it goes live.
Capabilities
Benchmarks
bar = score (0–100) · ▏ medianScores from vendor reports and public leaderboards. No private evals.
Showcase
Three fixed prompts, identical for every chat model. Outputs are archived verbatim — copy the prompt to reproduce.
Build a single-file HTML page using three.js from a CDN: a low-poly planet with a tilted ring of ~200 orbiting particles, a soft key light plus rim light, slow auto-rotation, and drag-to-orbit controls. No build tools — one file only.
Every model will run this exact prompt; the output lands here verbatim, with tokens and per-run cost.
Find the sum of all positive integers n such that n² + 85n + 2017 is a perfect square. Show your reasoning, then give only the final answer on the last line.
Every model will run this exact prompt; the output lands here verbatim, with tokens and per-run cost.
Extract strict JSON {vendor, invoice_no, issue_date, currency, total, line_items:[{desc, qty, unit_price}]} from the invoice text. Output JSON only.
"Northwind Cloud Ltd. — Invoice NW-20847, issued 12 Mar 2026. 3× GPU-hour bundle at $18.40, 1× object storage 2TB at $9.90. Total $65.10 (USD)."Every model will run this exact prompt; the output lands here verbatim, with tokens and per-run cost.
Quick Start
import os from openai import OpenAI client = OpenAI( base_url="https://api.mirapi.ai/v1", # changed api_key=os.environ["MIRAPI_API_KEY"], # changed ) resp = client.chat.completions.create( model="deepseek/deepseek-v3.2", messages=[{"role": "user", "content": "Hello"}], )
import OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://api.mirapi.ai/v1", // changed apiKey: process.env.MIRAPI_API_KEY, // changed }); const res = await client.chat.completions.create({ model: "deepseek/deepseek-v3.2", messages: [{ role: "user", content: "Hello" }], });
curl https://api.mirapi.ai/v1/chat/completions \ -H "Authorization: Bearer $MIRAPI_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek/deepseek-v3.2", "messages": [{"role": "user", "content": "Hello"}] }'
# ~/.zshrc export ANTHROPIC_BASE_URL="https://api.mirapi.ai" export ANTHROPIC_AUTH_TOKEN="$MIRAPI_API_KEY" # start as usual — model names unchanged claude --model deepseek/deepseek-v3.2
# ~/.codex/config.toml model_provider = "mirapi" model = "deepseek/deepseek-v3.2" [model_providers.mirapi] name = "MirAPI" base_url = "https://api.mirapi.ai/v1" env_key = "MIRAPI_API_KEY" wire_api = "responses"
Links
1M free tokens to start · no card · OpenAI-compatible, one base_url change