DeepSeek V4.1-Flash just landed in Command Code, and the launch numbers above (Terminal-Bench 3.0, DeepSWE v1.1, CyberGym, Automation-Bench) are agentic-coding scores, not a game build. To see what it does on a real one-shot UI task, we ran it against the same pipe-dodging bird flap game prompt this series has already put GLM 5.3 and Kimi K3 through - the "FlappyBench" round from our earlier 29-model comparison.
The prompt
Create a complete single HTML file for a cute bird flap game using Canvas. Smooth gravity, flapping animation, parallax background, score, pipe obstacles, game over screen
One shot each with /design. No corrections, no re-rolls.
Scorecard
Design + gameplay, scored 1-10 by hands-on play:
| Model | Score | Cost | Quality/$ |
|---|---|---|---|
| DeepSeek V4.1-Flash | 9/10 | $0.0089 | ~1,011 |
| GLM 5.3 | 9.5/10 | $0.0184 | ~516 |
| Kimi K3 | 8/10 | $0.0740 | ~108 |
DeepSeek V4.1-Flash one-shot a clean build at roughly half of GLM 5.3's cost and a ninth of Kimi K3's, landing just behind GLM 5.3 on feel. Kimi K3's build hard-codes parts of the gameplay rather than deriving them, which is the concrete gap behind its lower score here - a different finding from the earlier full read, where Kimi K3's code matched GLM 5.3 almost feature-for-feature. Scores are hands-on impressions from this round, not a re-run of that line-by-line audit.
Seen on X
This round was posted with the actual clip:
Ran FlappyBench on DeepSeek V4.1-Flash, GLM 5.3, and Kimi K3 with the same /design prompt. 🔹 DSV4.1-Flash: 9/10 · $0.0089 one-shot with the lowest cost 🔹 GLM-5.3: 9.5/10 · $0.0184 · nailed the UI but cost 2x more 🔹 Kimi K3: 8/10 · $0.0740 · hard-coded gameplay and 8x more...
Where this sits in the wider FlappyBench field
DeepSeek V4.1-Flash's $0.0089 is the cheapest of this three-model round, but it isn't the cheapest build this prompt has ever seen. From the full 29-model field:
| Model | Cost/gen |
|---|---|
| GLM 5.3 Flash | ~$0.0043 |
| DeepSeek V4.1-Flash | $0.0089 |
| GLM 5.3 | $0.0184 |
| Kimi K3 | $0.0740 |
| Gemini 3.8 Flash | ~$0.205 (multi-turn) |
GLM 5.3 Flash still undercuts DeepSeek V4.1-Flash by about half. Gemini 3.8 Flash sits at the other extreme - its ~$0.205 isn't a per-token story so much as an iteration-count one, needing several follow-up prompts to land a build this brief usually takes one shot to finish.
Which model should you reach for
If cost-per-shot is the deciding factor and you've already ruled out GLM 5.3 Flash, DeepSeek V4.1-Flash is the pick: a 9/10 one-shot at under a cent. If you want the single best-feeling build in this round and can spend roughly 2x more, GLM 5.3 edges it out. Kimi K3's hard-coded gameplay here is the one build in this round worth a follow-up prompt before shipping it.
That's the value of running the same prompt through every model instead of picking one on reputation: /design in Command Code makes swapping models a /model command, not a rewrite.
Try this yourself
1npm i -g command-code1/model deepseek-v4.1-flash
2/design Create a complete single HTML file for a cute bird flap game using Canvas. Smooth gravity, flapping animation, parallax background, score, pipe obstacles, game over screenGet the full comparison and code: CommandCodeAI/slash-design-showcase. Full command reference: commandcode.ai/docs/slash-commands/design.
