Cost per coding-agent request

Is GPT-6 Luna or Grok 4.7 cheaper for coding agents?

For coding agents, Grok 4.7 costs 35x as much as GPT-6 Luna per request ($0.13 vs $0.0038 on our measured traffic), because Grok 4.7 lists at $2 input, $0.50 cached input and $6 output per million tokens against $0.10, $0.01 and $0.50 for GPT-6 Luna, and both charge more for long prompts.

List prices: OpenAI API pricing, xAI models and pricing, checked 2026-10-13. Request costs: Standard Compute gateway data, 79,301 coding-agent requests from 75 users, Sep 13 to Oct 12, 2026. How we compute this

Run GPT-6 Luna in your agent from $19/month

See plans

Cost per coding-agent request, side by side

Cost at list price	GPT-6 Luna	Grok 4.7
Prompt under 100K tokens50% of requests, 87% of input cached	$0.0015	$0.044
Prompt of 100K to 272K39% of requests, 88% of input cached	$0.0039	$0.16
Prompt over 272K tokens11% of requests, 86% of input cached	$0.015	$0.48
Average request	$0.0038	$0.13
100 requests	$0.38	$13.16
Median session16 requests; a session ends after 10 idle minutes	$0.042	$1.21
A month of regular usemedian of 20 users active on 5+ days	$5.83	$222

Each logged request priced at the vendor’s list rates for its own token counts, including cache reads and any higher rate for long prompts. Average output: 698 tokens per request. Both models are priced on the requests up to 500K tokens that fit in both context windows; band shares above describe all requests.

Benchmarks, speed and latency, side by side

Artificial Analysis	GPT-6 Luna	Grok 4.7
Intelligence Index10 evaluations, 0 to 100	38	46
Terminal-Bench 4.0agentic coding in a terminal	12.6%	25.8%
Output speedmedian	136 tokens/s	86 tokens/s
Time to first answer tokenincludes thinking time	101 s	101 s

Source: Artificial Analysis (artificialanalysis.ai). Measured variants: GPT-6 Luna (Max), speed on OpenAI; Grok 4.7 (Xhigh), speed on SpaceXAI. Fetched October 13, 2026. Artificial Analysis runs Terminal-Bench 4.0 in its own setup, so its scores differ from the tbench.ai leaderboard.

What to watch for

Grok 4.7's 500K window cannot take 1% of coding-agent requests, so the comparison uses only requests both can take. 87% of agent input is read from cache, and Grok 4.7 charges $0.50 per million for it against $0.01. Grok 4.7 scores clearly higher on the benchmarks above, so it may finish hard tasks in fewer steps.

FAQ

Is GPT-6 Luna cheaper than Grok 4.7 for coding agents?

Yes, at list prices. 100 average coding-agent requests cost $0.38 on GPT-6 Luna and $13.16 on Grok 4.7. A median 16-request session costs $0.042 vs $1.21. Measured on 79,301 real coding-agent requests (Sep 13 to Oct 12, 2026), counting only the requests up to 500K tokens that both can take.

How much does a month of Grok 4.7 vs GPT-6 Luna cost?

For the median regular user in our data (1,478 requests in 30 days), about $222 on Grok 4.7 and $5.83 on GPT-6 Luna at list price. Heavier users pay proportionally more.

GPT-6 Luna vs Grok 4.7: which is better for agents?

Grok 4.7 scores higher on output quality (8.6/10 vs 7.5/10), while GPT-6 Luna is the cheaper option. GPT-6 Luna is best for cheap, fast subtasks, routing and subagents in OpenAI-based stacks; Grok 4.7 is best for output-heavy agent work with prompts under 200K tokens. Live community votes on this page show what people actually prefer.

Can I switch between GPT-6 Luna and Grok 4.7 in the same agent?

Yes. Model-agnostic agents (Cline, Aider, OpenClaw, Hermes Agent, Kilo Code, OpenCode, Continue) let you swap models per task. Many people route hard tasks to the stronger model and routine work to the cheaper one.