What are the verified coding and reasoning benchmarks for GPT-5.4?

GPT-5.4 achieves SWE-bench Verified: 76.8% (Thinking), LiveCodeBench: 80.1%, AIME: 95.2%. Verified sources: OpenAI GPT-5.4 Architecture Milestone (OpenAI).

How do I run GPT-5.4 with Swarm?

Configure GPT-5.4 via CLI using: swarm ctl model set --provider codex --model codex/gpt-5.4, or switch models interactively inside the terminal TUI with /models.

Does Swarm isolate GPT-5.4 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-5.4 be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure GPT-5.4 as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

GPT-5.4 — Specs, Context & Capabilities

Q: What is the recommended Swarm multi-agent topology for GPT-5.4?

Swarm recommends GPT-5.4 as Primary / Planner (Tier 1: Master Autonomous Architect). Supreme orchestration capabilities for decomposing complex epics into deterministic Task Programs. Best paired with: Gemini 3.8 Flash for subagent implementation.

GPT-5.4 Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

76.8%

ThinkingOpenAI Technical Report ↗

LiveCodeBench

80.1%

ThinkingLiveCodeBench ↗

LMSYS Coding Elo

1445Elo

Coding ArenaLMSYS Arena ↗

AIME 2024 / 2025

95.2%

ThinkingOpenAI Technical Report ↗

Pricing ($ / 1M tokens)

$2.5/ $10

$0.65 cached readStandard API Pricing

What This Means in Real Software Development

Sets new performance standards: 76.8% on SWE-bench Verified and 80.1% on LiveCodeBench.
95.2% on AIME 2024 demonstrates near-flawless multi-step algorithmic deduction.
Flawless handling of large repository schemas, database migrations, and distributed microservices.
Unrivaled tool calling stability during complex multi-checkpoint Swarm workflows.

Swarm Multi-Agent Topology

Tier 1: Master Autonomous ArchitectRole: Primary / Planner

Best Paired With: Gemini 3.8 Flash for subagent implementation

Supreme orchestration capabilities for decomposing complex epics into deterministic Task Programs.

Git Worktree Isolation Advantage:

Supervises multi-worktree execution and reviews integration diffs before promotion.

What Developers & Practitioners Are Saying

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

Field Consensus

"Advanced software synthesis engine bridging broad natural language understanding with rigorous type-safe code authoring."

Verified Real-World Strengths

76.8% on SWE-bench Verified
High-fidelity translation across disparate languages (e.g. Python to Go)
Stable multi-agent coordination

Production Caveats & Gaps

Requires prompt guidance to adhere strictly to project styling guidelines

Primary Sources & Methodology Citations

OpenAI GPT-5.4 Architecture Milestone — OpenAI (March 2026)

GPT-5.4 model overview

GPT-5.4 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: 1M tokens
Thinking support: off · low · medium · high · xhigh
Availability status: Limited
Catalog ID: codex/gpt-5.4

Image input
Text input
Text output
Reasoning
Tools
Limited

Run GPT-5.4 in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider codex --model codex/gpt-5.4

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign GPT-5.4 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 →