Harness comparison
Codex vs DeepSeek Harness — DeepSeek Harness vs Codex: 4 side-by-side tests
4 posts from X ran OpenAI Codex and DeepSeek Harness on the same task, between Sep 24, 2026 and Oct 19, 2026. Models: GPT-6.1 Sol, DeepSeek V4.1 Flash 1017, Claude Opus 5.5, DeepSeek V4 Pro 0813, Gemini 3.1 Pro, GPT-6 Sol. All harness comparisons.
Posts
4
Tetris in three agents: Claude Code, Codex, DeepSeek: Claude Opus 5.5, GPT-6.1 Sol, DeepSeek V4 Pro 0813. Video by @agentsky_dev.
3D scene from a reference image: DeepSeek V4.1 Flash 1017, GPT-6.1 Sol. Video by @NFT_Chen.
Escape-the-cops game, 3 harnesses: Gemini 3.1 Pro, DeepSeek V4 Pro 0813, GPT-6 Sol. Video by @Furkan_popr.
Codex vs DeepSeek Harness, test by test
| Test | Codex: model | DeepSeek Harness: model | Time (Codex / DeepSeek Harness) | Author’s pick |
|---|---|---|---|---|
| Nesting dolls in Three.js, Codex vs DeepSeek Harness@yangyang_2048 | GPT-6.1 Sol (xhigh) | DeepSeek V4.1 Flash 1017 | — | — |
| Tetris in three agents: Claude Code, Codex, DeepSeek@agentsky_dev | GPT-6.1 Sol | DeepSeek V4 Pro 0813 | 3.4 min / 3.4 min | — |
| 3D scene from a reference image@NFT_Chen | GPT-6.1 Sol (high) | DeepSeek V4.1 Flash 1017 | — | DeepSeek Harness |
| Escape-the-cops game, 3 harnesses@Furkan_popr | GPT-6 Sol (Medium) | DeepSeek V4 Pro 0813 (High) | — | OpenAI Codex |
Numbers are as each post states them; “—” means the post gives none. Tests and models differ from row to row.
Codex vs DeepSeek Harness 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-6.1 Sol | Codex | $2prompts up to 272K tokens | $10prompts up to 272K tokens |
| GPT-6 Sol | Codex | ||
| DeepSeek V4.1 Flash 1017 | DeepSeek Harness | $0.30peak hours; off-peak is half price | $1.20peak hours; off-peak is half price |
| DeepSeek V4 Pro 0813 | DeepSeek Harness |
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
In every test below each side ran its own model, so these are not controlled comparisons of the harnesses alone.
Nesting dolls in Three.js, Codex vs DeepSeek Harness — GPT-6.1 Sol (xhigh) in Codex vs DeepSeek V4.1 Flash 1017 in DeepSeek Harness
Same Three.js task: four nesting dolls open, line up from largest to smallest, then nest back together. GPT-6.1 Sol (xhigh) in Codex on the left, DeepSeek V4.1 Flash 1017 in DeepSeek Harness on the right.
- OpenAI Codex with GPT-6.1 Sol at xhigh effort — Left.
- DeepSeek Harness with DeepSeek V4.1 Flash 1017 — Right.
Posted Oct 19, 2026 by Yangyang (@yangyang_2048). Original post on X · Video and details
Tetris in three agents: Claude Code, Codex, DeepSeek — GPT-6.1 Sol in Codex vs DeepSeek V4 Pro 0813 in DeepSeek Harness
Same Tetris prompt in three agents: Claude Code with Claude Opus 5.5 (3:58, $0.60), Codex with GPT-6.1 Sol (3:25, $0.16) and DeepSeek Harness with DeepSeek V4 Pro (3:24, $0.26). All three games are playable. The post also includes Claude Code.
- Claude Code with Claude Opus 5.5: time 4 min — Playable; its AI clears lines.
- OpenAI Codex with GPT-6.1 Sol: time 3.4 min — Playable; cheapest of the three.
- DeepSeek Harness with DeepSeek V4 Pro 0813: time 3.4 min — Playable, but its turn ended mid self-test.
As stated: Claude Code 3:58, $0.60; Codex 3:25, $0.16; DeepSeek Harness 3:24, $0.26.
Posted Oct 19, 2026 by AgentSky (@agentsky_dev). Original post on X · Video and details
3D scene from a reference image — GPT-6.1 Sol (high) in Codex vs DeepSeek V4.1 Flash 1017 in DeepSeek Harness
DeepSeek V4.1 Flash 1017 in DeepSeek Harness (right) and GPT-6.1 Sol in Codex at high effort (left) build a 3D scene from the same reference image. The author prefers DeepSeek’s warm dusk lighting and richer detail.
- DeepSeek Harness with DeepSeek V4.1 Flash 1017 — Right. Author’s pick.
- OpenAI Codex with GPT-6.1 Sol at high effort — Left.
Verdict from @NFT_Chen: DeepSeek harness clearly understands looks and atmosphere better in this 3D round (paraphrased).
Posted Oct 18, 2026 by SuSu (@NFT_Chen). Original post on X · Video and details
Escape-the-cops game, 3 harnesses — GPT-6 Sol (Medium) in Codex vs DeepSeek V4 Pro 0813 (High) in DeepSeek Harness
A run-from-the-cops game from one-shot prompts in three harness and model pairs: Google Antigravity with Gemini 3.1 Pro, DeepSeek Harness with DeepSeek V4 Pro, and Codex with GPT-6 Sol. The post also includes Google Antigravity.
- Google Antigravity with Gemini 3.1 Pro at High effort — Free, per the author.
- DeepSeek Harness with DeepSeek V4 Pro 0813 at High effort — About $2, per the author.
- OpenAI Codex with GPT-6 Sol at Medium effort — Codex on a $20 subscription. Author’s pick.
Verdict from @Furkan_popr: Only one of the three is actually playable (paraphrased).
Note: The author marks the Antigravity and DeepSeek Harness runs with a clown emoji, implying the Codex run is the playable one.
Posted Sep 24, 2026 by Furkan (@Furkan_popr). Original post on X · Video and details
Codex vs DeepSeek Harness (DeepSeek Harness vs Codex): FAQ
Is Codex better than DeepSeek Harness?
Author picks: 2 posts of 4 name a winner between the two — OpenAI Codex 1, DeepSeek Harness 1. Each post is a different task, so watch the outputs above for the kind of work you do.
Which is faster, Codex or DeepSeek Harness?
1 post states a time for both: DeepSeek Harness was faster in it. Times are as each author states them; models and settings differ.
Which is cheaper, Codex or DeepSeek Harness?
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.
Which models were used with Codex and DeepSeek Harness?
With Codex: GPT-6.1 Sol, GPT-6 Sol. With DeepSeek Harness: DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro 0813.