What are the verified coding and reasoning benchmarks for GPT-6-Luna?

GPT-6-Luna achieves SWE-bench Verified: 64.5% (max), LiveCodeBench: 56.8%. Verified sources: Artificial Analysis Intelligence Index v4.3.2 (Artificial Analysis), OpenAI GPT-6 Luna Efficiency Report (OpenAI).

What is the recommended Swarm multi-agent topology for GPT-6-Luna?

Swarm recommends GPT-6-Luna as Coder Subagent (Tier 2: Extreme Efficiency Subagent). Handles dozens of parallel subtasks at near-zero incremental cost. Best paired with: gpt-6-astra or claude-opus-5.

How does GPT-6-Luna compare to rival frontier coding models on SWE-bench and pricing?

GPT-6-Luna achieves SWE-bench Verified: 64.5% at $0.12/$0.48 per 1M tokens. Direct peer comparison: Qwen 3.8 27B (62% SWE-bench, $0.2/$0.8); GLM 5.3 Flash (66.8% SWE-bench, $0.3/$1.2); K2 Horizon 375B A23B (59.8% SWE-bench, $0.4/$1.6); OpenAI o4-mini (56.8% SWE-bench, $1.2/$4.8).

How do I run GPT-6-Luna with Swarm?

Configure GPT-6-Luna via CLI using: swarm ctl model set --provider codex --model codex/gpt-6-luna, or switch models interactively inside the terminal TUI with /models.

Does Swarm isolate GPT-6-Luna 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-6-Luna be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure GPT-6-Luna as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

GPT-6-Luna — Specs, Context & Capabilities

GPT-6-Luna Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

64.5%

maxOpenAI Lab ↗

LiveCodeBench

56.8%

ThinkingLiveCodeBench ↗

LMSYS Coding Elo

1358Elo

Coding ArenaLMSYS Arena ↗

Pricing ($ / 1M tokens)

$0.12/ $0.48

$0.02 cached readStandard API Pricing

What This Means in Real Software Development

Achieves the lowest cost per Intelligence task globally at an unbelievable $0.07/task.
Outputs 157 tokens per second for lightning-fast subagent iterations.
Scores 37 on the Artificial Analysis Intelligence Index v4.3.2 and 39 on Finance & Accounting.
Priced at just $0.12/M input and $0.48/M output tokens.

Swarm Multi-Agent Topology

Tier 2: Extreme Efficiency SubagentRole: Coder Subagent

Best Paired With: gpt-6-astra or claude-opus-5

Handles dozens of parallel subtasks at near-zero incremental cost.

Git Worktree Isolation Advantage:

Ideal for high-fanout Coder worktree waves running concurrent test fixes.

What Developers & Practitioners Are Saying

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

Field Consensus

"OpenAI's most economical reasoning model ever created. At $0.07 per task, teams run continuous automated unit test generation and lint repair 24/7."

Verified Real-World Strengths

Absolute lowest cost per task ($0.07) in the Artificial Analysis Index
High-throughput 157 TPS streaming rate
Reliable OpenAI tool use and schema validation

Production Caveats & Gaps

Intelligence Index (37) below Tier 1 flagships for deep architectural design
Requires supervision from GPT-6 Astra on complex multi-stage tasks

Primary Sources & Methodology Citations

Artificial Analysis Intelligence Index v4.3.2 — Artificial Analysis (September 2026)
OpenAI GPT-6 Luna Efficiency Report — OpenAI (June 2026)

Head-to-Head Frontier Comparisons: GPT-6-Luna vs Rivals

Standardized, side-by-side benchmark & economics comparison against peer models

Flagship Comparative Matrix

Model	Role / Tier	SWE-bench	LiveCodeBench	LMSYS Elo	AIME	Input / Output
GPT-6-LunaOpenAI Codex (OAuth)Current	Tier 2: Extreme Efficiency Subagent	64.5%	56.8%	1358	—	$0.12 / $0.48
Qwen 3.8 27BQwen Compare GPT-6-Luna vs Qwen 3.8 27B →	Tier 2: Open-Weights Specialist	62%	54.5%	1342	—	$0.2 / $0.8
GLM 5.3 FlashZhipu AI Compare GPT-6-Luna vs GLM 5.3 Flash →	Tier 2: High-Speed Coder Subagent	66.8%	58.5%	1365	—	$0.3 / $1.2
K2 Horizon 375B A23BIndependent Compare GPT-6-Luna vs K2 Horizon 375B A23B →	Tier 2: MoE Systems Specialist	59.8%	52%	1330	—	$0.4 / $1.6
OpenAI o4-miniOpenAI Compare GPT-6-Luna vs OpenAI o4-mini →	Tier 2: High-Speed Deductive Engine	56.8%	73.4%	1385	90.1%	$1.2 / $4.8

Explore All AI Model Comparisons Hub (GPT-6-Luna & Flagship Peers) →

GPT-6-Luna model overview

GPT-6-Luna 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 6
Context window: 1.05M tokens
Thinking support: off · low · medium · high · xhigh · max
Availability status: Available
Catalog ID: codex/gpt-6-luna

Image input
Text input
Text output
Reasoning
Tools

Run GPT-6-Luna in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider codex --model codex/gpt-6-luna

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign GPT-6-Luna 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 →