<Frame caption="The Arena sign-in screen">
<img src="/images/arena-sign-in.jpg" alt="DualMind Arena sign-in screen with email, phone, GitHub and Google options" />
</Frame>
</Step>
<Step title="Stay in Battle mode" icon="swords">
The Arena opens in **Battle** mode: two anonymous models, picked at random. You can switch to **Side by Side** (you choose both models) or **Direct Chat** (one model) from the mode menu at the top. See [Arena modes](/concepts/model-comparison).
<Frame caption="The mode menu at the top of the Arena">
<img src="/images/arena-modes.jpg" alt="Arena mode menu listing Battle, Side by Side and Direct Chat" style={{ maxWidth: "320px" }} />
</Frame>
</Step>
<Step title="Send a prompt" icon="message-square">
Type a prompt and press Enter. Both models answer the same prompt at the same time, labelled **Model A** and **Model B**.
<Tip>
Open-ended prompts reveal more than trivia. Try: *"Explain the difference between TCP and UDP to a five-year-old using a pizza delivery analogy."*
</Tip>
</Step>
<Step title="Vote" icon="vote">
Read both answers, then choose **Left is better**, **Right is better**, **It's a tie** or **Both are bad**. One vote per battle; it can't be changed.
<Frame caption="Two anonymous answers and the vote bar. Sample answers shown.">
<img src="/images/arena-battle.jpg" alt="Arena battle screen with Model A and Model B answers and four vote buttons" />
</Frame>
</Step>
<Step title="See who you picked" icon="eye">
After your vote the model names are revealed and your vote counts toward the [leaderboard](/concepts/leaderboard). Keep going in the same thread, or [share it](/concepts/threads) with a link.
</Step>
</Steps>
<Check>
That's it: you've run a blind battle and your vote is on the leaderboard.
</Check>
The Arena runs on a public REST API at `{{apiBase}}`. Reading the leaderboard needs no account:
<CodeGroup>
```bash cURL icon="terminal"
curl {{apiBase}}/api/arena/model-stats
```
```javascript JavaScript
const res = await fetch("{{apiBase}}/api/arena/model-stats");
const { items } = await res.json();
console.table(items.slice(0, 5).map(m => ({ rank: m.eloRank, model: m.displayName, elo: m.eloScore })));
```
```python Python
import requests
items = requests.get("{{apiBase}}/api/arena/model-stats").json()["items"]
for m in items[:5]:
print(m["eloRank"], m["displayName"], m["eloScore"])
```
</CodeGroup>
Battles and votes need a signed-in user. Grab your access token ([how](/api-reference/authentication)), then start a battle:
<CodeGroup>
```bash cURL
curl {{apiBase}}/api/arena/dualchat \
-H "Authorization: Bearer $DUALMIND_TOKEN" \
-H "Content-Type: application/json" \
-d '{"prompt": "Write a haiku about distributed systems", "selectionMode": "random"}'
```
```javascript JavaScript
const res = await fetch("{{apiBase}}/api/arena/dualchat", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.DUALMIND_TOKEN}`,
"Content-Type": "application/json",
},
body: JSON.stringify({ prompt: "Write a haiku about distributed systems", selectionMode: "random" }),
});
const battle = await res.json();
console.log(battle.comparisonId, battle.agent1.message, battle.agent2.message);
```
```python Python
import os, requests
battle = requests.post(
"{{apiBase}}/api/arena/dualchat",
headers={"Authorization": f"Bearer {os.environ['DUALMIND_TOKEN']}"},
json={"prompt": "Write a haiku about distributed systems", "selectionMode": "random"},
timeout=120,
).json()
print(battle["comparisonId"], battle["agent1"]["message"], battle["agent2"]["message"])
```
</CodeGroup>
Then vote with the <Tooltip headline="comparisonId" tip="The ID of this battle, returned by POST /api/arena/dualchat. You vote on it with POST /api/arena/model-vote." cta="Submit a vote" href="/api-reference/voting/submit-vote">`comparisonId`</Tooltip>. The [Run a battle](/api-reference/guides/run-a-battle) tutorial walks through the whole flow.