What are the verified coding and reasoning benchmarks for GPT 5.2?

GPT 5.2 achieves SWE-bench Verified: 73.5% (Autonomous Agent Mode), LiveCodeBench: 77%, AIME: 93.5%. Verified sources: OpenAI GPT-5.2 Codex Technical Report (OpenAI).

How do I run GPT 5.2 with Swarm?

Configure GPT 5.2 via CLI using: swarm ctl model set --provider openai --model openai/gpt-5.2, or switch models interactively inside the terminal TUI with /models.

Does Swarm isolate GPT 5.2 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 5.2 be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure GPT 5.2 as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

GPT 5.2 — Specs, Context & Capabilities

Q: What is the recommended Swarm multi-agent topology for GPT 5.2?

Swarm recommends GPT 5.2 as Coder Subagent (Tier 1: Dedicated Systems Coder). High-fidelity Coder worker for intricate codebases requiring zero build regression. Best paired with: Claude 3.7 Sonnet as Lead Planner.

GPT 5.2 Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

73.5%

Autonomous Agent ModeOpenAI Codex System Card ↗

LiveCodeBench

77%

ThinkingLiveCodeBench ↗

LMSYS Coding Elo

1430Elo

Coding ArenaLMSYS Arena ↗

AIME 2024 / 2025

93.5%

ThinkingOpenAI Codex System Card ↗

Pricing ($ / 1M tokens)

$2.5/ $10

$0.6 cached readStandard API Pricing

What This Means in Real Software Development

73.5% autonomous SWE-bench score with reduced hallucination in complex asynchronous code.
Writes idiomatic Go, Rust, and TypeScript with strict adherence to repository linter configurations.
Operates with responsive execution, maintaining high developer productivity during live agent runs.
Native support for Swarm Task Program stage barriers and verification handoffs.

Swarm Multi-Agent Topology

Tier 1: Dedicated Systems CoderRole: Coder Subagent

Best Paired With: Claude 3.7 Sonnet as Lead Planner

High-fidelity Coder worker for intricate codebases requiring zero build regression.

Git Worktree Isolation Advantage:

Executes isolated feature work inside dedicated worktrees with full Git isolation.

What Developers & Practitioners Are Saying

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

Field Consensus

"Refined developer-centric Codex engine, optimized for low-latency interactive code editing."

Verified Real-World Strengths

73.5% on SWE-bench Verified
Fast AST traversal and refactoring suggestions
Superb unit test generation coverage

Production Caveats & Gaps

Can be aggressive with dependency version upgrades if not constrained

Primary Sources & Methodology Citations

OpenAI GPT-5.2 Codex Technical Report — OpenAI (December 2025)

GPT 5.2 model overview

GPT 5.2 is listed in the public Swarm catalog through OpenAI. This page summarizes the verified catalog fields used by Swarm without claiming real-time access for every provider account.

Provider: OpenAI
Family: Gpt 5
Context window: 400K tokens
Thinking support: off · low · medium · high · xhigh
Availability status: Available
Catalog ID: openai/gpt-5.2

File input
Image input
Text input
Text output
Reasoning
Tools
Structured output
JSON
Batch

Run GPT 5.2 in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider openai --model openai/gpt-5.2

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign GPT 5.2 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 →