GPT-6 Luna vs DeepSeek V4 Flash: is Luna now cheaper?

The short answer

OpenAI lists GPT-6 Luna at $0.10 per million input tokens, $0.01 per million cached input tokens, and $0.50 per million output tokens for standard API processing with short context. DeepSeek lists Flash at $0.15 uncached input and $0.60 output during off-peak hours. On those two rates, Luna costs 33% less for input and about 17% less for output. [1, 2]

Price per million tokens

USD per 1M text tokens. Luna column: OpenAI Standard, short context. DeepSeek columns: official Flash cache-hit, cache-miss, and output rates. DeepSeek peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday, excluding Chinese public holidays; other times are off-peak. [1, 2]

DeepSeek is the cheaper choice for cache hits: $0.003 off-peak or $0.006 peak, versus Luna's $0.01 cached input. Luna also charges $0.125 per million cache-write tokens, a separate line item in OpenAI's price card. A workload with extensive reusable context can therefore produce a different winner from the headline input/output comparison. [1, 2]

A coding-workload example

Imagine 1 million uncached input tokens and 200,000 output tokens across a batch of coding tasks. At the listed token rates, Luna costs $0.20; DeepSeek Flash costs $0.27 off-peak or $0.54 at peak. That is roughly 26% less than the off-peak DeepSeek bill for this specific mix.

This is an illustrative token calculation, not an estimate of requests or coding tasks included in a plan. Real jobs differ in input/output ratio, cache-hit rate, reasoning-token use, retries, tool calls, and how much work each model finishes successfully. The cheaper model per token is not automatically cheaper per completed task.

What about the Codex plan?

There is a second way to use Luna: OpenAI says GPT-6 Luna is rolling out in Codex, including Plus, Pro, Business, Enterprise, and Edu access; Free and Go users can access Luna in the desktop app. Rollout and workspace settings affect availability. That is useful if you already code in Codex, but it is separate from buying Luna through the API. A Codex subscription does not turn the published $0.10/$0.50 API rates into an unlimited token allowance, and its usage limits depend on the plan. [3]

If by “coding plan” you mean a bundled subscription rather than the OpenAI Codex product, compare that plan's actual model access, request limits, and reset windows separately. An API rate card alone cannot tell you which subscription offers more coding sessions.

Which should you try?

Try Luna if your workload is mostly uncached text input and output, you want a low published API price at any hour, or you already have access in Codex. Try DeepSeek Flash if your workflow gets a high cache-hit rate, or if its responses complete your particular coding tasks with fewer iterations. Benchmark both on the same prompts and count the full bill per successful task.

One more guardrail: OpenAI doubles Luna's input and cache rates, and raises output rates by 50%, for requests exceeding 272,000 input tokens. Regional processing and other processing modes also have different prices. DeepSeek has its own peak/off-peak schedule. Always check the live rate cards before committing to a budget. [1, 2]