Reasoning Models

Models that think before they answer.

Chain-of-thought models for tasks that require planning, math, science, code, and multi-step logic. Same API as chat, smarter results.

Chain-of-thought

Models think through problems step by step. Higher accuracy on math, logic, and code.

Tool use

Reasoning models can call functions, browse, and execute code. Multi-step agent loops out of the box.

128K context

Reason over entire codebases, papers, or books. Citations to source material.

Self-verification

Models check their own work. Higher accuracy on high-stakes outputs.

Same API

Drop-in replacement. Reasoning is just another model parameter.

Streaming tokens

Watch the model think in real time. Inspect intermediate steps for debugging.

Pricing

Token-efficient, with volume discounts and burn incentives.

Reason-1 (fast)
0.0024QUBIC / 1K tokens

P50 1.8s, 32K context

Reason-1 (deep)
0.0080QUBIC / 1K tokens

P50 6s, 128K context

Reason-1 (max)
0.024QUBIC / 1K tokens

P50 24s, 256K context

Tool calls
0.0001QUBIC / call

Plus underlying token cost

Pricing is illustrative. Final rates are governed by on-chain parameters and may vary based on network state.

Staking requirements

Tier-based access. Higher stakes unlock better economics and more capacity.

TierRequired stakeAccess
Builder50M QUBICAll reasoning models, best-effort
Startup150M QUBICReserved throughput, deep + max
Business500M QUBICCustom reasoning models
EnterpriseCustomOn-prem, private models

Example

Drop-in compatible with the OpenAI SDK.

reasoning.py
from aigarth import Aigarth

client = Aigarth(api_key="sk-aigarth-...")

response = client.chat.create(
    model="aigarth-reason-1-deep",
    messages=[
        {"role": "user", "content": "A train leaves Boston at 9am at 60mph. Another leaves NYC at 10am at 80mph. When do they meet?"},
    ],
)

# Inspect the reasoning
for step in response.choices[0].message.reasoning_steps:
    print(step)
print("Answer:", response.choices[0].message.content)

Enterprise benefits

Everything in the standard tier, plus the things enterprises need.

  • Best-in-class accuracy on math and reasoning benchmarks
  • Self-verification reduces hallucinations in production
  • Citations and source attribution for every answer
  • Tool use without separate orchestration SDK
  • Streaming tokens let users see the model think
  • Used by Helix Labs, Vector Capital, Lumen Legal
  • Migration from o1, o3-mini, Claude with extended thinking
  • On-prem deployment for sensitive workloads

Ready to get started?

Open the console, generate an API key, and run your first call in minutes.