Cost per coding-agent request

Is Grok 4.20 or Qwen3 235B A22B cheaper for coding agents?

For coding agents, Grok 4.20 costs 4.5x as much as Qwen3 235B A22B per request ($0.044 vs $0.0096 on our measured traffic), because Grok 4.20 lists at $1.25 input, $0.20 cached input and $2.50 output per million tokens against $0.09, $0.09 and $0.55 for Qwen3 235B A22B, and Grok 4.20 charges more for prompts 200K tokens or more.

List prices: xAI models and 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 current Claude, GPT, Gemini and open models in your agent

Or let the router send each step to the cheapest model that can do it, on providers we vet for price, uptime and speed. Plans from $19 a month.

See plans

Cost per coding-agent request, side by side

Cost at list price	Qwen3 235B	Grok 4.20
Prompt under 100K tokens50% of requests, 87% of input cached	$0.0055	$0.021
Prompt of 100K to 272K39% of requests, 88% of input cached	$0.015	$0.074
Prompt over 272K tokens11% of requests, 86% of input cached	n/a	n/a
Average request	$0.0096	$0.044
100 requests	$0.96	$4.37
Median session16 requests; a session ends after 10 idle minutes	$0.091	$0.42
A month of regular usemedian of 20 users active on 5+ days	$13.05	$60.65

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	Qwen3 235B	Grok 4.20
Intelligence Index10 evaluations, 0 to 100	12 (estimate)	26 (estimate)
Output speedmedian	48 tokens/s	95 tokens/s
Time to first answer tokenincludes thinking time	1.3 s	21 s

Source: Artificial Analysis (artificialanalysis.ai). Measured variants: Qwen3 235B A22B 2507 Instruct; Grok 4.20 0309 v2 (Reasoning), 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

Qwen3 235B A22B's 262K window cannot take 12% of coding-agent requests, so the comparison uses only requests both can take. The price is OpenRouter's listed rate on October 13, 2026, which can be a third-party host's; Alibaba's own API may charge differently. Grok 4.20 scores clearly higher on the benchmarks above, so it may finish hard tasks in fewer steps.

FAQ

Is Qwen3 235B A22B cheaper than Grok 4.20 for coding agents?

Yes, at the listed prices. 100 average coding-agent requests cost $0.96 on Qwen3 235B A22B and $4.37 on Grok 4.20. A median 16-request session costs $0.091 vs $0.42. 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 Grok 4.20 vs Qwen3 235B cost?

For the median regular user in our data (1,478 requests in 30 days), about $60.65 on Grok 4.20 and $13.05 on Qwen3 235B A22B at list price. Heavier users pay proportionally more.

Grok 4.20 vs Qwen3 235B A22B: which is better for agents?

Grok 4.20 scores higher on output quality (5.8/10 vs 3.8/10), while Qwen3 235B A22B is the cheaper option. Grok 4.20 is best for runs that need more than 1M tokens of context; 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 Grok 4.20 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.

Output Quality

Correctness and depth of what it produces

Agentic Ability

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

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 run on 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: Opt-in rules keep a company, team or key 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 Stripe4,100+ agent users