What are the verified coding and reasoning benchmarks for GPT-5.3 Codex Spark?

GPT-5.3 Codex Spark achieves SWE-bench Verified: 75% (Fast Autonomous Mode), LiveCodeBench: 78.6%, AIME: 94.2%. Verified sources: GPT-5.3 Codex Spark System Performance (OpenAI).

Does Swarm isolate GPT-5.3 Codex Spark code edits from other agents?

Yes. Swarm provisions isolated Git worktrees for every session and subagent, preventing file race conditions and keeping your main working tree clean until verification passes.

Can GPT-5.3 Codex Spark be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure GPT-5.3 Codex Spark as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

GPT-5.3 Codex Spark — Specs, Context & Capabilities

Q: What is the recommended Swarm multi-agent topology for GPT-5.3 Codex Spark?

Swarm recommends GPT-5.3 Codex Spark as Coder Subagent (Tier 1: High-Speed Frontier Coder). High throughput and elite coding logic make it ideal for rapid iterative development cycles. Best paired with: Claude 3.7 Sonnet as Lead Planner.

Q: How does Swarm isolate GPT-5.3 Codex Spark during autonomous code generation?

Executes rapid code edits and commits to isolated Git worktrees without blocking the user.

Q: What are developers and practitioners saying about GPT-5.3 Codex Spark in production?

Ultra-responsive Codex variant tailored for instant completions and rapid subagent triage.

Q: How do I run GPT-5.3 Codex Spark with Swarm?

Configure GPT-5.3 Codex Spark via CLI using: swarm ctl model set --provider codex --model codex/gpt-5.3-codex-spark, or switch models interactively inside the terminal TUI with /models.

GPT-5.3 Codex Spark Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

75%

Fast Autonomous ModeCodex Technical Report ↗

LiveCodeBench

78.6%

ThinkingLiveCodeBench ↗

LMSYS Coding Elo

1438Elo

Coding ArenaLMSYS Arena ↗

AIME 2024 / 2025

94.2%

ThinkingCodex Technical Report ↗

Pricing ($ / 1M tokens)

$2.5/ $10

$0.45 cached readStandard API Pricing

What This Means in Real Software Development

75.0% on SWE-bench Verified with rapid execution.
Optimized for real-time terminal interactions, shell command generation, and fast build troubleshooting.
Near-zero first-token latency (<300ms TTFT) makes it feel instantaneous in the Swarm TUI.
Highly stable for automated test repair loops and continuous integration pipelines.

Swarm Multi-Agent Topology

Tier 1: High-Speed Frontier CoderRole: Coder Subagent

Best Paired With: Claude 3.7 Sonnet as Lead Planner

High throughput and elite coding logic make it ideal for rapid iterative development cycles.

Git Worktree Isolation Advantage:

Executes rapid code edits and commits to isolated Git worktrees without blocking the user.

What Developers & Practitioners Are Saying

Real-world field observations, production strengths, and developer consensus

Field Consensus

"Ultra-responsive Codex variant tailored for instant completions and rapid subagent triage."

Verified Real-World Strengths

High throughput
Instant start-up time for fast terminal command validation
Affordable token economics

Production Caveats & Gaps

Less suitable for deep architectural system planning

Primary Sources & Methodology Citations

GPT-5.3 Codex Spark System Performance — OpenAI (February 2026)

GPT-5.3 Codex Spark model overview

GPT-5.3 Codex Spark is listed in the public Swarm catalog through OpenAI Codex (OAuth). This page summarizes the verified catalog fields used by Swarm without claiming real-time access for every provider account.

Provider: OpenAI Codex (OAuth)
Family: Gpt 5
Context window: 128K tokens
Thinking support: off · low · medium · high
Availability status: Available
Catalog ID: codex/gpt-5.3-codex-spark

Image input
Text input
Text output
Reasoning
Tools

Run GPT-5.3 Codex Spark in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider codex --model codex/gpt-5.3-codex-spark

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign GPT-5.3 Codex Spark to specialized system roles in ~/.swarm/config.json:

Primary / Planner: Orchestrates full task DAGs and multi-stage pipelines.
Coder Subagent: Generates code and unit tests inside private Git worktrees.
Finder Subagent: Performs read-only code search, symbol discovery, and mapping.

System Agent Roles →Why Swarm Architecture →