Codex vs Pi — Pi vs Codex: 4 side-by-side tests
4 posts from X ran OpenAI Codex and Pi on the same task, between Oct 19, 2026 and Oct 28, 2026. Models: GPT-6 Astra, Gemini 4 Argon, Claude Opus 5.5, GPT-6.1 Terra, GPT-6.1 Sol, GPT-6 Luna. All harness comparisons.
- 4 posts state a time for both: OpenAI Codex was faster in all 4, median gap ~2.1×.
- 4 posts state a cost for both, split evenly (OpenAI Codex 2, Pi 2).
- 1 of 4 keeps the model fixed (GPT-6 Luna); in the rest each side ran its own model, so harness and model effects are mixed.
From 4 posts; costs and times as stated by each author, not a controlled benchmark.
Posts
4
Supercar drive, three labs' agents on day one: GPT-6.1 Terra, Claude Opus 5.5, Gemini 4 Argon. Posted by @agentsky_dev.
Ink-wash landscape, three agents blind guess: GPT-6.1 Sol, Gemini 4 Argon, Claude Opus 5.5. Posted by @agentsky_dev.
GLASS CATHEDRAL fracture benchmark, four harnesses: GPT-6 Luna. Posted by @AlicanKiraz0.
Codex vs Pi, test by test
| Test | Codex: model | Pi: model | Time (Codex / Pi) | Cost (Codex / Pi) | Author’s pick |
|---|---|---|---|---|---|
| 3D volcano island, Astra vs Argon vs Opus@agentsky_dev | GPT-6 Astra | Gemini 4 Argon | 3.9 min / 4.7 min | $0.47 / $0.44 | — |
| Supercar drive, three labs' agents on day one@agentsky_dev | GPT-6.1 Terra | Gemini 4 Argon | 3.5 min / 8.7 min | $0.12 / $0.48 | — |
| Ink-wash landscape, three agents blind guess@agentsky_dev | GPT-6.1 Sol | Gemini 4 Argon | 106 s / 4.6 min | $0.08 / $0.52 | — |
| GLASS CATHEDRAL fracture benchmark, four harnesses@AlicanKiraz0 | GPT-6 Luna (High) | GPT-6 Luna (High) | 24 min / 42 min | $0.34 / $0.15 | — |
Numbers are as each post states them; “—” means the post gives none. Tests and models differ from row to row.
Codex vs Pi pricing: plans
From each vendor’s official pricing page (checked 2026-10-19); prices exclude tax and can change. Usage limits differ by plan and model.
API list prices of the models in these tests
| Model | Used with | Input $/1M tokens | Output $/1M tokens |
|---|---|---|---|
| GPT-6 Astra | Codex | $10 | $50 |
| GPT-6.1 Terra | Codex | ||
| GPT-6.1 Sol | Codex | $2prompts up to 272K tokens | $10prompts up to 272K tokens |
| GPT-6 Luna | Codex, Pi | ||
| Gemini 4 Argon | Pi | $2introductory price through Nov 30, 2026; $4 after | $10introductory price through Nov 30, 2026; $20 after |
Official API list prices where verified; a blank cell has no verified official price in the archive (older or retired models).
What each test showed
Unless a test says both sides ran the same model, each side ran its own model — not a controlled comparison of the harnesses alone.
3D volcano island, Astra vs Argon vs Opus — GPT-6 Astra in Codex vs Gemini 4 Argon in Pi
GPT-6 Astra in Codex, Gemini 4 Argon in pi and Claude Opus 5.5 in Claude Code get one real task after the index reshuffle: a 3D volcano island. The post also includes Claude Code.
- OpenAI Codex with GPT-6 Astra: time 3.9 min, cost $0.47.
- Pi with Gemini 4 Argon: time 4.7 min, cost $0.44.
- Claude Code with Claude Opus 5.5: time 11.4 min, cost $1.71.
As stated: Astra 3 min 51 s, $0.47; Argon 4 min 40 s, $0.44; Opus 5.5 11 min 27 s, $1.71.
Posted Oct 28, 2026 by AgentSky (@agentsky_dev). Post on X · Video and details
Supercar drive, three labs' agents on day one — GPT-6.1 Terra in Codex vs Gemini 4 Argon in Pi
Same supercar prompt on GPT-6.1 Terra's first day, each lab's own agent: Codex with GPT-6.1 Terra, Claude Code with Claude Opus 5.5 and pi with Gemini 4 Argon, labelled on screen. The post also includes Claude Code.
- OpenAI Codex with GPT-6.1 Terra: time 3.5 min, cost $0.12.
- Claude Code with Claude Opus 5.5: time 8.3 min, cost $1.31.
- Pi with Gemini 4 Argon: time 8.7 min, cost $0.48 — Second try.
As stated: Codex (GPT-6.1 Terra) 3 min 27 s, $0.12; Claude Code (Claude Opus 5.5) 8 min 21 s, $1.31; pi (Gemini 4 Argon) 8 min 41 s, $0.48, second try.
Note: The poster asks which car you would drive.
Posted Oct 27, 2026 by AgentSky (@agentsky_dev). Post on X · Video and details
Ink-wash landscape, three agents blind guess — GPT-6.1 Sol in Codex vs Gemini 4 Argon in Pi
Codex with GPT-6.1 Sol, pi with Gemini 4 Argon and Claude Code with Opus 5.5 get the same ink-wash landscape prompt, shown as A, B and C before a reveal with time and cost. The post also includes Claude Code.
- OpenAI Codex with GPT-6.1 Sol: time 106 s, cost $0.08.
- Pi with Gemini 4 Argon: time 4.6 min, cost $0.52.
- Claude Code with Claude Opus 5.5: time 7.6 min, cost $1.20.
As stated: Codex (GPT-6.1 Sol) 1 min 46 s, $0.08; pi (Gemini 4 Argon) 4 min 36 s, $0.52; Claude Code (Opus 5.5) 7 min 36 s, $1.20.
Posted Oct 26, 2026 by AgentSky (@agentsky_dev). Post on X · Video and details
GLASS CATHEDRAL fracture benchmark, four harnesses — GPT-6 Luna (High) in Codex vs GPT-6 Luna (High) in Pi
GPT-6 Luna at high effort runs the GLASS CATHEDRAL fracture benchmark in Pi, OMP, Droid and Codex CLI: three Gothic stained-glass windows, 150–200 breakable glass fragments with XPBD constraints, in one HTML file. Both sides ran GPT-6 Luna, so this one isolates the harness. The post also includes Droid and OMP.
- Pi with GPT-6 Luna at High effort: time 42 min, cost $0.15 — Score 6.43/10.
- Droid with GPT-6 Luna at High effort: time 15 min, cost $0.17 — Score 9.33/10. Author’s pick.
- OpenAI Codex with GPT-6 Luna at High effort: time 24 min, cost $0.34 — Score 5.86/10.
- OMP with GPT-6 Luna at High effort: time 1 h 17 min, cost $0.83 — Score 3.10/10.
As stated: Droid $0.17, 15 min; Pi $0.15, 42 min; Codex CLI $0.34, 24 min; OMP $0.83, 1 h 17 min.
Verdict from @AlicanKiraz0: Droid stood out by combining speed with low cost (9.33/10); Pi was cheapest, Codex CLI second fastest, and OMP ranked last (paraphrased).
Note: Scores out of 10 from GPT-6 Astra Pro as judge LLM. Planning-first, then development, in every tool.
Posted Oct 19, 2026 by Alican Kiraz (@AlicanKiraz0). Post on X · Video and details
Codex vs Pi (Pi vs Codex): FAQ
Is Codex better than Pi?
None of the 4 posts names a winner between the two.
Which is faster, Codex or Pi?
OpenAI Codex, going by the posts' own numbers: it was faster in 4 of the 4 posts that give a time for each side (median gap ~2.1×). Times are as each author states them; models and settings differ.
Which is cheaper, Codex or Pi?
Neither, going by the posts' own numbers: the 4 posts that give a cost for each side come out even (OpenAI Codex 2, Pi 2). Those are the run costs the posts state. Codex plans (official pricing page, checked 2026-10-19): Free $0/month, Go $8/month, Plus $20/month, Pro From $100/month, Business $25/user/month, API key API pricing.
Did any test run the same model in Codex and Pi?
Yes, in 1 of the 4: GLASS CATHEDRAL fracture benchmark, four harnesses (@AlicanKiraz0) ran GPT-6 Luna on both sides. The others pair each harness with a different model.
Which models were used with Codex and Pi?
With Codex: GPT-6 Astra, GPT-6.1 Terra, GPT-6.1 Sol, GPT-6 Luna. With Pi: Gemini 4 Argon, GPT-6 Luna.