Cost per coding-agent request

Is GPT-5.6 Luna or Qwen3 235B A22B cheaper for coding agents?

For coding agents, Qwen3 235B A22B costs 1.9x as much as GPT-5.6 Luna per request ($0.0096 vs $0.0052 on our measured traffic), because Qwen3 235B A22B lists at $0.09 input, $0.09 cached input and $0.55 output per million tokens against $0.20, $0.02 and $1.20 for GPT-5.6 Luna, and GPT-5.6 Luna charges more for prompts over 272K tokens.

List prices: OpenAI API pricing, OpenRouter listed price, 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-5.6 Luna and GPT-6 Luna in your agent from $19/month

See plans

Cost per coding-agent request, side by side

Cost at list price	GPT-5.6 Luna	Qwen3 235B
Prompt under 100K tokens50% of requests, 87% of input cached	$0.0032	$0.0055
Prompt of 100K to 272K39% of requests, 88% of input cached	$0.0078	$0.015
Prompt over 272K tokens11% of requests, 86% of input cached	n/a	n/a
Average request	$0.0052	$0.0096
100 requests	$0.52	$0.96
Median session16 requests; a session ends after 10 idle minutes	$0.063	$0.091
A month of regular usemedian of 20 users active on 5+ days	$6.53	$13.05

Each logged request priced at the listed rates above 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 262K tokens that fit in both context windows; band shares above describe all requests.

Benchmarks, speed and latency, side by side

Artificial Analysis	GPT-5.6 Luna	Qwen3 235B
Intelligence Index10 evaluations, 0 to 100	37	12 (estimate)
Terminal-Bench 4.0agentic coding in a terminal	11.6%	n/a
Output speedmedian	123 tokens/s	48 tokens/s
Time to first answer tokenincludes thinking time	117 s	1.3 s

Source: Artificial Analysis (artificialanalysis.ai). Measured variants: GPT-5.6 Luna (Max), speed on OpenAI; Qwen3 235B A22B 2507 Instruct. 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

Qwen3 235B A22B's 262K window cannot take 12% of coding-agent requests, so the comparison uses only requests both can take. 87% of agent input is read from cache, and Qwen3 235B A22B charges $0.09 per million for it against $0.02. GPT-5.6 Luna is superseded: GPT-6 Luna costs $0.0041 per request.

FAQ

Is GPT-5.6 Luna cheaper than Qwen3 235B A22B for coding agents?

Yes, at the listed prices. 100 average coding-agent requests cost $0.52 on GPT-5.6 Luna and $0.96 on Qwen3 235B A22B. A median 16-request session costs $0.063 vs $0.091. Measured on 79,301 real coding-agent requests (Sep 13 to Oct 12, 2026), counting only the requests up to 262K tokens that both can take.

How much does a month of Qwen3 235B vs GPT-5.6 Luna cost?

For the median regular user in our data (1,478 requests in 30 days), about $13.05 on Qwen3 235B A22B and $6.53 on GPT-5.6 Luna at list price. Heavier users pay proportionally more.

GPT-5.6 Luna vs Qwen3 235B A22B: which is better for agents?

GPT-5.6 Luna scores higher on output quality (7.4/10 vs 3.8/10), while Qwen3 235B A22B is the cheaper option. GPT-5.6 Luna is best for cheap subtasks in existing GPT-5.6 setups; Qwen3 235B A22B is best for self-hosted setups that need a mid-size open-weight model. Live community votes on this page show what people actually prefer.

Can I switch between GPT-5.6 Luna and Qwen3 235B A22B in the same agent?

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

Community vote

Pick a winner in each dimension. You can change your vote any time. 3 votes cast so far.

Output Quality

Correctness and depth of what it produces

100%2 votes0%

Agentic Ability

Tool calls, instruction-following, and multi-step tasks

0%1 vote100%

Speed

Tokens per second and time-to-first-token

Value for Money

How much capability you get per dollar

Reliability

Consistent results: fewer refusals, loops and format breaks

Current Claude, GPT, Gemini and open models are included with Standard Compute

Standard Compute

Current frontier models Claude Fable 5.1 · GPT-6 Astra

More intelligence
per dollar.

Each step goes to the cheapest model that can do it, on providers we vet for price, uptime and speed, with automatic fallback when one fails.

Compliance: Keep a key, team or company on European or US providers
Privacy: We don’t train on your prompts or store them

Get your API key See how it works

Fixed monthly plans from $19 · cancel anytime

7-day fair refund Free trial, no card Billing by Stripe3,400+ agent users