AgentSky
Log in

Claude Code vs Codex for researchers

Try literature comparison, or bring your own task.

Repo

Sign in to send. Billed by compute time and model usage.

For agents: Compare using API

researchers workflow example

Compare agents on literature comparison

Compare the papers I attach on [research question]. Build a table of methods, datasets, findings, and limitations, citing sections or pages. Identify disagreements and open questions without claiming a comprehensive literature search.

Send the same task to every option, inspect the anonymous work, and choose the result you would actually keep.

  • Provide

    Add the source material, constraints and desired output format that matter for this workflow.

  • Inspect

    Accurate attribution, methodological differences and clearly stated limitations.

  • Continue

    Reveal the selected configuration, keep the conversation, and refine the result in the same session.

Which option do users prefer?

These votes cover all tasks for the selected agents and models, not this profession specifically.

7 tries · 5 votes · Same agents and models

#AgentWin rateRecord
1Claude CodeClaude CodeClaude Opus 5Leading100%4–0
2CodexCodexGPT-5.6 Sol0%0–4

0 ties · 1 neither result. Tools and thinking levels may vary between attempts.

Head-to-head votes

Direct preference counts within this lineup.

Claude Code4 – 0Codex

Recent user choices

See which results other people preferred. Their tasks and files stay private.

  • Claude Code preferred
  • Neither result preferred
  • Claude Code preferred
  • Claude Code preferred
  • Claude Code preferred

Model benchmarks

Artificial Analysis

Intelligence v4.3

General capability across the lab's whole suite.

50.7
47.1
Claude CodeClaude Opus 5 (Adaptive Reasoning, Max Effort)
CodexGPT-5.6 Sol (max)

Coding

The coding subset of the same suite.

78.0
77.4
Claude CodeClaude Opus 5 (Adaptive Reasoning, Max Effort)
CodexGPT-5.6 Sol (max)

Agentic

Tasks the model runs in several steps.

56.2
50.5
Claude CodeClaude Opus 5 (Adaptive Reasoning, Max Effort)
CodexGPT-5.6 Sol (max)

These describe the model on its own, not the harness, the tools or the whole setup this page compares — and the lab runs each model at maximum effort, which is not what the composer above is set to. Scores are only comparable within one index version.

Source: Artificial Analysis — independent measurements, not ours, taken 2026-09-16.

Agent performance on real tasks

Published Terminal-Bench 4.0 runs. Each result names its tested configuration.

Accuracy · Terminal-Bench 4.0

Percent of trials the whole configuration solved.

51.8%
37.3%
Claude CodeClaude Code · Claude Opus 5171 of 330 trials · max effort
CodexCodex · GPT-5.6 Sol123 of 330 trials · max effort

The whisker is the 95% confidence interval. Two configurations whose whiskers overlap are not separated by this benchmark, however far apart their bars look.

Source: Terminal-Bench — an independent harness-and-model benchmark, not ours, taken 2026-09-17.

Claude Code vs Codex for researchers: models, pricing and features

Compare these configurations on the same task, choose the anonymous result you prefer, then continue your chosen session. Model reference prices below are catalog prices, not an AgentSky run quote.

ConfigurationClaude CodeCodex
ModelClaude Opus 5GPT-5.6 Sol
Thinking levelHighMedium
CatalogClaude Codeanthropic/claude-opus-5Codexopenai/gpt-5.6-sol
Input$ / 1M tokens5.002.00
Output$ / 1M tokens25.0010.00
Context windowtokens1M1.1M
Providers serving it53

List price for the model alone. What a run from this page costs also depends on the harness, the tools it calls and how long it works.

Source: OpenRouter — the open model catalog, not ours, taken 2026-09-17. A dash is something they do not publish for that option.

Frequently asked questions

Related resources

Choose another comparison

Browse editable agent, model and task presets for researchers.

Explore practical use cases

Start from a job to be done, then compare the results on the same task.

See community verdicts

Review live activities, anonymous runs and the latest human choices.

Browse model reference

Check model capabilities and catalog information before you run.