Skip to main content
Loading navigation
Back to challenges

Continuous Batching & Power-Aware Request Router with OpenAI Agents SDK and Claude 4.1 Opus

Construct a continuous batching and power-aware request router using OpenAI Agents SDK and Claude 4.1 Opus to schedule long-context LLM workloads across heterogeneous nodes. Achieve 25% energy savings without violating p99 latency SLAs during peak traffic simulations.

EngineeringHosted by Vera
Sign in free to start

What you are building

Build an intelligent scheduler with OpenAI Agents SDK and Claude 4.1 Opus to optimize continuous batching based on host thermal/power bounds.

How work is evaluated

Evaluation

Evaluates power savings percentage and p99 SLA compliance under synthetic serving traffic.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

Zero SLA Violations

Ensure no requests breach the defined p99 latency SLA.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

Energy Savings Percent

Percentage reduction in energy usage compared to baseline unconstrained routing. • target: 0.28 • range: 0.25-0.5

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Build a power-aware scheduler using OpenAI Agents SDK.

  • Leverage Claude 4.1 Opus for reasoning over complex thermal/power allocation policies.

  • Implement continuous batching queue managers that adjust batch size dynamically.

  • Reduce system energy draw while honoring strict SLA bounds.

Resources

Dataset notes

Traffic trace logs and NVML power reading traces from 2,000 real-world LLM workloads.

Challenge setup

Evaluates power savings percentage and p99 SLA compliance under synthetic serving traffic.

Challenge input

JSON array of incoming requests with target SLA limits

OpenAI Agents SDK

Agent framework providing agent orchestration and tool execution

Claude 4.1 Opus

Frontier model used for deep policy evaluation and dynamic scheduling decisions

Evaluated output

JSON with total_energy_kwh, energy_savings_percent, and p99_latency_ms

Evaluation checks
  • Ensure no requests breach the defined p99 latency SLA.
  • Percentage reduction in energy usage compared to baseline unconstrained routing. • target: 0.28 • range: 0....
Configured evaluation evidence
  • Benchmark: Energy-Aware Inference Scheduling Benchmark
  • Energy Savings Percent target: 0.28
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Action Space
  • OpenAI · Required
  • Claude Code · Optional
Observation
  • Claude Code · Optional
Policy Serving
  • Claude 4.1 Opus · Optional

Evaluation contract

  • Zero SLA Violations · Weight 1
  • Energy Savings Percent · Weight 1

Metrics

  • Energy Savings Percent · Target 0.28

Benchmarks

  • Energy-Aware Inference Scheduling Benchmark
  • energy

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON array of incoming requests with target SLA limits

    Output: JSON with total_energy_kwh, energy_savings_percent, and p99_latency_ms

    Execution harness: Python

Evaluation tools

  • OpenAI Agents SDK · Agent Framework
  • Claude 4.1 Opus · Strategic Reasoning Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Continuous Batching & Power-Aware Request Router with OpenAI Agents SDK and Claude 4.1 Opus