Virtual poolant:reasoning

ANT Reasoning

Thinking-capable models only, with a quality floor of 70.

Extended thinking: RequiredQuality floor: 70/100 or above

How this pool routes

ant:reasoning hard-requires models that support extended thinking, then weights quality heavily. Non-reasoning models are excluded no matter how well they score, so multi-step problems get a model that can actually deliberate.

Best For

Multi-step reasoningMath and logicPlanning

Where requests went

Live routing over the last 30 days

Requests

Success rate

98.36%

Avg latency

888ms

qwen/qwen3-next-80b-a3b-thinkingSemantic-cache83.6% · 51

openai/o3-miniOpenRouter13.1% · 8

ant:reasoningUnknown1.6% · 1

openai/gpt-5OpenRouter1.6% · 1

Models in this pool

39 models across 10 providers currently qualify. Per-request limits (context size, cost caps, your model allowlist) narrow this further at routing time.

Showing the 24 highest-scoring of 39.

Quick Start

from openai import OpenAI

client = OpenAI(
    base_url="https://api.antbase.ai/v1",
    api_key="YOUR_ANT_API_KEY",
)

response = client.chat.completions.create(
    model="ant:reasoning",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

Routing weights

Speed30/100

Quality90/100

Cost30/100

Selection criteria

Extended thinkingrequiredRequired

Quality floorrequired70/100 or above

Speed weight30/100

Quality weight90/100

Cost weight30/100

Model IDant:reasoning

Other pools

Every virtual pool accepts the same request shape — swap the model id to change the routing policy.

Browse all virtual models →