OpenAI

Codex: Coding Agent Index Score, Performance & Price

Compared with other coding agents, Codex is above average on the Coding Agent Index, but somewhat expensive. It's also notably fast. Its best result uses GPT-6 Astra (Max), and we tested it with 10 models.

Codex with GPT-6 Astra (Max) scores 55 on the Artificial Analysis Coding Agent Index, placing it above average among coding agents (median: 49).

At 24.8m per task, Codex with GPT-6 Astra (Max) is notably fast (median: 36.1m).

On average, it costs $7.04 in API usage per task to run Codex with GPT-6 Astra (Max) on the Coding Agent Index (somewhat expensive, median: $5.41).

Performance

Performance across the Artificial Analysis Coding Agent Index.

Artificial Analysis Coding Agent Index

Artificial Analysis Coding Agent Index v2.0 incorporates 3 benchmarks: FrontierCode v1.1, Terminal-Bench 4.0, and SWE-Atlas-QnA-AA · Higher is better

Color by

Router models

The Artificial Analysis Coding Agent Index is a composite score built from FrontierCode v1.1, Terminal-Bench 4.0, and SWE-Atlas-QnA-AA.

It is useful for quick comparison, but it should be read alongside the per-eval breakdowns. Two agents with similar index values can still have different strengths across repository tasks, terminal workflows, and rubric-based evaluations.

Artificial Analysis Coding Agent Index vs. Cost per Task

Artificial Analysis Coding Agent Index vs. average pay-per-token API cost per task (USD)

Color by

Router models

Most attractive quadrant

Pareto line

Each point represents a coding-agent variant. Farther left means lower average cost per task, while higher on the chart means higher benchmark performance. The most efficient agents sit toward the upper-left: stronger results at lower cost.

Token Usage

Token consumption across the Artificial Analysis Coding Agent Index, including total usage, token mix, efficiency, and per-benchmark breakdowns.

Token Usage per Task

Average input, cache, and output tokens per task

Router models

Prompt cache hit rates can vary significantly by provider routing, which can materially change effective cost.

Non-cached input tokens sent to the model, including prompts, instructions, tool context, and task context that were not served from prompt cache.

Artificial Analysis Coding Agent Index vs. Total Tokens

Artificial Analysis Coding Agent Index vs. average total tokens per task

Color by

Router models

Most attractive quadrant

Each point represents a coding-agent variant. Farther left means lower average total token usage per task, while higher on the chart means higher benchmark performance. Agents toward the upper-left achieve stronger results with fewer tokens.

Cost

Cost across the Artificial Analysis Coding Agent Index based on current per-token API pricing, including cache write pricing and cache discounts where available. Many users will access coding agent harnesses through subscription plan offerings rather than pay-per-token.

Cost per Task

Average pay-per-token API cost per task (USD) · Lower is better

Color by

Router models

This chart shows the average pay-per-token API cost per task across the Artificial Analysis Coding Agent Index, spanning FrontierCode v1.1, Terminal-Bench 4.0, and SWE-Atlas-QnA-AA.

Where applicable, that cost model includes standard input pricing, discounted cached-input pricing, separate cache-write charges, and output pricing rather than treating all prompt tokens as if they were billed at the same uncached input rate.

It is intended to show pay-per-token API cost, not consumer plan pricing or the full operational cost of deploying the system in production. Infrastructure, engineering, and supervision costs are not the focus of this metric.

Execution Time

Active agent runtime across the Artificial Analysis Coding Agent Index.

Time per Task

Average agent wall time per task · Lower is better

Color by

Router models

This chart uses agent wall time: how long the agent process was actively running on each task.

It does not include environment startup, verifier or judge time, or other harness overhead, so it is a cleaner comparison of how long the agent itself was working.

Artificial Analysis Coding Agent Index vs. Execution Time

Artificial Analysis Coding Agent Index vs. average agent wall time per task

Color by

Router models

Most attractive quadrant

Each point represents a coding-agent variant. Farther left means shorter average agent runtime per task, while higher on the chart means higher benchmark performance. Agents toward the upper-left deliver stronger results in less active agent time.

Safety Refusals

Methodology

How often a provider or model refuses to proceed with a task on safety grounds across the Artificial Analysis Coding Agent Index.

Safety Refusal Rate

Share of retained task attempts with a safety refusal · Equal weight per benchmark · Lower is better

Break down by

Router models

Safety and fallback behaviour is provider-configurable and may change over time; these rates reflect behaviour recorded at the time of benchmarking.

Blocked ended the attempt. Fallback means another model finished it. Both count towards the refusal rate.

Reward Hacking

Methodology

Share of task attempts that would otherwise pass but are scored zero following an integrity violation.

Reward Hacking Rate: FrontierCode v1.1

Share of task attempts scored zero for reward hacking · Uses a modified version of Cognition's rule-based scanner to detect prohibited access to solution material · Lower is better

Color by

Router models

Codex: Coding Agent Index Score, Performance & Price

Comparison Summary

Performance

Artificial Analysis Coding Agent Index

What This Metric Means

Artificial Analysis Coding Agent Index vs. Cost per Task

How to Read This Chart

Token Usage

Token Usage per Task

Input Tokens

Artificial Analysis Coding Agent Index vs. Total Tokens

How to Read This Chart

Cost

Cost per Task

What Cost Is Measuring

Execution Time

Time per Task

What Execution Time Is Measuring

Artificial Analysis Coding Agent Index vs. Execution Time

How to Read This Chart

Safety Refusals

Safety Refusal Rate

Safety Refusal Outcomes

Reward Hacking

Reward Hacking Rate: FrontierCode v1.1

Run Specifications