by OpenAICoding AgentfreemiumRank #4 on Leaderboard

Codex

Coding agent harness from OpenAI, measured on the Artificial Analysis Coding Agent Index.

Overall Index
53.1/ 100
Artificial Analysis benchmark composite
SWE-bench Verified
—
Real-world GitHub issue resolution
FrontierCode-AA 2.0
42.6%
Multi-file software engineering tasks
Terminal-Bench 4.1
57.6%
Shell execution & command accuracy
Avg Cost / Task
$7.63
Inference token expenditure per task

Overview & Analysis

Codex was evaluated by Artificial Analysis with GPT-6 Astra (Max). We have not written a full profile for it yet.

Pricing Model: See the vendor for current pricing.

Technical Specs & Models

Primary EngineGPT-6 Astra (Max)
EnvironmentNot recorded
Platforms
LicenseNot recorded
Custom API Keys (BYOK)Bundled Only (No)
Supported Foundation Models:
GPT-6 Astra (Max)

Key Capabilities & Features

Specialized engineering mechanisms implemented in Codex.

Pros & Cons Evaluation

Forensic analysis of developer experience, cost efficiency, and trade-offs.

Strengths & Advantages
    Limitations & Trade-Offs

      Pricing Plans & Tiers

      Subscription plans, credit quotas, and commercial deployment options.

      Verified Benchmark Evaluations

      Standardized evaluation scores recorded on SWE-bench and synthetic code harnesses.

      Benchmark HarnessScore / AccuracyEvaluation MetricVerification StatusEvaluation Notes
      Coding Agent Index (v2.1)53.1%Composite of three benchmarksVerifiedSource: Artificial Analysis.
      FrontierCode-AA 2.042.6%Software engineering tasksVerifiedStandard benchmark instance run.
      Terminal-Bench 4.157.6%Agentic terminal useVerifiedStandard benchmark instance run.
      SWE-Atlas-QnA-AA59.1%Technical Q&AVerifiedStandard benchmark instance run.