DeepSeek V4 Pro 0813 Max vs GPT-6 Luna Max compared

DeepSeek V4 Pro 0813 (Max) from DeepSeek and GPT-6 Luna (Max) from OpenAI are both tracked in Franklin AI's benchmark catalogue. The sharpest difference is price: GPT-6 Luna Max costs $0.50 per million output tokens, against $3.96 for DeepSeek V4 Pro 0813 Max. Everything below is read from the benchmark, price and speed data in Franklin AI's catalogue; nothing here comes from a hands-on test.

What they cost

DeepSeek V4 Pro 0813 Max lists $1.32 per million input tokens and $3.96 per million output tokens; GPT-6 Luna Max lists $0.10 and $0.50. That makes GPT-6 Luna Max's output 87% cheaper.

For a sample workload of 1 million input tokens and 200,000 output tokens, DeepSeek V4 Pro 0813 Max would cost $2.11 and GPT-6 Luna Max $0.20, a difference of $1.91. A bigger example, summarising 1,000 documents of 8,000 tokens into 400-token summaries, costs $12.14 on DeepSeek V4 Pro 0813 Max and $1.00 on GPT-6 Luna Max: a gap of $11.14. On a 3:1 blend of input to output tokens, that works out to $1.98 versus $0.20 per million.

When they came out

GPT-6 Luna Max arrived on September 22, 2026, 40 days after DeepSeek V4 Pro 0813 Max (August 13, 2026). They are effectively contemporaries.

How fast they respond

Once a response is under way, DeepSeek V4 Pro 0813 Max produces roughly 67 tokens per second against 121 tokens per second for GPT-6 Luna Max. That makes GPT-6 Luna Max roughly 1.8 times as fast at writing. Time to first token is 1.7 seconds for DeepSeek V4 Pro 0813 Max and 2.0 minutes for GPT-6 Luna Max. GPT-6 Luna Max writes faster once it starts, but DeepSeek V4 Pro 0813 Max's head start is large enough that DeepSeek V4 Pro 0813 Max still finishes first at any typical reply length; GPT-6 Luna Max only catches up beyond about 18,000 tokens. Waits that long are felt on every chat-style request, though they matter less for batch jobs.

Put together, a 1,000-token answer would take about 17 seconds from DeepSeek V4 Pro 0813 Max and 2.2 minutes from GPT-6 Luna Max, counting the first-token wait.

The headline benchmark

On the intelligence index, DeepSeek V4 Pro 0813 Max scores 23.1 against 27.9 for GPT-6 Luna Max. That is a clear margin for an index that averages many evaluations.

Among 109 active scored models, DeepSeek V4 Pro 0813 Max sits 72nd and GPT-6 Luna Max 62nd.

For coding or maths work specifically, the catalogue has no dedicated index on either side yet.

Score per dollar

Dividing the blended price by the intelligence index gives a rough cost per point: $0.086 for DeepSeek V4 Pro 0813 Max and $0.0072 for GPT-6 Luna Max, per million tokens. GPT-6 Luna Max scores higher and costs less per point, so on price per point there is nothing to trade off.