Grok 4.7 lists at $2.00 input, $6.00 output and $0.50 cached input per 1M tokens, with a 500,000-token context window and reasoning that is always on. Grok 4.3 lists at $1.25 input, $2.50 output and $0.20 cached input, with a 1,000,000-token context window and reasoning on by default. In my data, Grok 4.7 costs more per task than Grok 4.3 on all four tasks, and Grok 4.3 is not the cheapest option on any of them.
Grok list prices and what they mean
Grok pricing works like most API models: you pay per million input tokens, per million output tokens, and a lower rate for cached input. Reasoning tokens are billed as output tokens, so a model that thinks a lot costs more even when its visible answer is short.
These are the list prices and settings I used, as of 2026-10-22:
| Model | Input / Output / Cached input (per 1M) | Context | Reasoning mode | LMArena Text |
|---|---|---|---|---|
| Grok 4.7 | $2.00 / $6.00 / $0.50 | 500,000 | always | 1444 |
| Grok 4.3 | $1.25 / $2.50 / $0.20 | 1,000,000 | default | 1442 |
| DeepSeek V4.1 Flash | $0.30 / $1.20 / $0.006 | 1,048,576 | default | 1475 |
| Gemini 3.5 Flash Lite | $0.30 / $2.50 / $0.03 | 1,048,576 | always | 1455 |
Grok 4.7 has the higher list price and the smaller context window. Both Grok models also use more input tokens on the same prompt than the others, which I come back to below.
Cost per task
Costs are per task, with the cost per 1,000 tasks in brackets. I measured Grok 4.7 and DeepSeek V4.1 Flash directly. Grok 4.3 and Gemini 3.5 Flash Lite are estimates. The agent task is modeled.
Chat answer
The median across all 334 models is 0.24¢.
- Grok 4.7: 0.96¢ ($9.58 per 1,000), measured; 1,280 in / 567 out / 603 reasoning tokens.
- Grok 4.3: 0.45¢ ($4.53 per 1,000), estimated; same token counts.
- DeepSeek V4.1 Flash: 0.76¢ ($7.60 per 1,000), measured; 65 in / 859 out / 5,461 reasoning tokens.
- Gemini 3.5 Flash Lite: 0.41¢ ($4.09 per 1,000), estimated; 35 in / 788 out / 843 reasoning tokens.
Grok 4.3 and Gemini 3.5 Flash Lite are close. DeepSeek V4.1 Flash has a much lower list price than Grok 4.7 but ends up close to it per answer because it produced 5,461 reasoning tokens. The Grok runs also carry 1,280 input tokens against 35 to 65 for the others, so the prompt size affects the comparison.
| # | Model | Cost per task | Text score | Tokens in / out | Basis | |
|---|---|---|---|---|---|---|
| 1 |  GLM 5.3 FlashZ.ai | 0.02¢$0.22 / 1,000 | 1474 | 48 / 1,782 | measured best value | |
| 2 |  GPT-6 LunaOpenAI | 0.06¢$0.59 / 1,000 | 1443 | 45 / 1,179 | measured | |
| 3 |  Claude Haiku 5.5Anthropic | 0.07¢$0.66 / 1,000 | - | 61 / 1,299 | measured | |
| 4 |  Qwen3.8 FlashAlibaba Qwen | 0.08¢$0.76 / 1,000 | - | 88 / 1,598 | measured | |
| 5 |  Mistral Small 4Mistral | 0.08¢$0.83 / 1,000 | - | 40 / 1,377 | estimated | |
| 6 |  MiMo-V2.6-ProXiaomi | 0.09¢$0.90 / 1,000 | 1479 | 206 / 933 | estimated best value |
Blog post
The median for a 1,500-word post is 0.80¢.
- Grok 4.7: 2.3¢ ($22.66 per 1,000), measured; 1,296 in / 3,117 out / 227 reasoning tokens.
- Grok 4.3: 1.00¢ ($9.98 per 1,000), estimated; same token counts.
- DeepSeek V4.1 Flash: 0.37¢ ($3.67 per 1,000), measured; 81 in / 2,230 out / 809 reasoning tokens.
- Gemini 3.5 Flash Lite: 0.90¢ ($8.98 per 1,000), estimated; 52 in / 2,621 out / 964 reasoning tokens.
DeepSeek V4.1 Flash is the cheapest here. Grok 4.3 and Gemini 3.5 Flash Lite are close to each other, and Grok 4.7 is the most expensive of the four.
Classify 100 support tickets
The median is 0.68¢.
- Grok 4.7: 2.2¢ ($22.30 per 1,000), measured; 2,512 in / 479 out / 2,400 reasoning tokens.
- Grok 4.3: 1.0¢ ($10.34 per 1,000), estimated; same token counts.
- DeepSeek V4.1 Flash: 0.63¢ ($6.30 per 1,000), measured; 1,296 in / 461 out / 4,462 reasoning tokens.
- Gemini 3.5 Flash Lite: 1.7¢ ($16.99 per 1,000), estimated; 1,397 in / 571 out / 6,059 reasoning tokens.
Here Gemini 3.5 Flash Lite costs more than Grok 4.3 because of its 6,059 reasoning tokens. Both Grok models sit above the median.
50-step coding agent
This is a modeled scenario: an 8,000-token system prompt, 50 steps, 1,500 extra context tokens per step and 300 visible output tokens per step. The median is 48¢.
- Grok 4.7: $1.70 ($1,702.19 per 1,000), or $5.18 without caching.
- Grok 4.3: 73¢ ($726.42 per 1,000), or $3.16 without caching.
- DeepSeek V4.1 Flash: 16¢ ($160.27 per 1,000), or $0.79 without caching.
- Gemini 3.5 Flash Lite: 24¢ ($242.10 per 1,000), or $0.81 without caching.
Caching matters most here. Grok 4.7 falls from $5.18 to $1.70 once cached input is counted. If you run agents, check caching first, whichever model you choose.
| # | Model | Cost per task | WebDev score | Tokens in / out | Basis | |
|---|---|---|---|---|---|---|
| 1 | GPT-6 LunaOpenAI | 4.1¢$40.71 / 1,000 | 1578 | 2,237,500 / 22,000 | modeled best value | |
| 2 | GLM 5.3 FlashZ.ai | 5.0¢$50.10 / 1,000 | 1606 | 2,239,737 / 43,814 | modeled best value | |
| 3 | MiMo-V2.6-ProXiaomi | 6.5¢$64.51 / 1,000 | 1625 | 2,241,975 / 24,380 | modeled best value | |
| 4 | Claude Haiku 5.5Anthropic | 6.6¢$66.35 / 1,000 | 1585 | 3,101,175 / 50,340 | modeled | |
| 5 | GPT-5.4 NanoOpenAI | 8.7¢$86.92 / 1,000 | 1401 | 2,237,500 / 22,000 | modeled | |
| 6 | Qwen3.8 FlashAlibaba Qwen | 9.2¢$92.31 / 1,000 | - | 2,259,875 / 96,000 | modeled |
When reasoning raises the bill
Reasoning tokens are billed as output, so the reasoning count can matter more than the list price. Examples from the data:
- Qwen3.8 Flash on the blog post used 16,000 reasoning tokens, hit the output cap and returned 0 visible output tokens. It still cost 0.75¢.
- Mistral Small 4 on the agent used 132,350 reasoning tokens and cost 13¢. GPT-5.4 Nano used 7,000 reasoning tokens and cost 8.7¢.
- DeepSeek V4.1 Flash on chat used 5,461 reasoning tokens, which pushed it to 0.76¢ despite a low list price.
In my data, Grok 4.3 and Grok 4.7 have identical reasoning counts on each task (603 on chat, 227 on the blog post, 2,400 on classification, 45,450 on the agent). Their cost gap therefore comes from list price, not from extra thinking. Grok 4.7 always reasons, so you cannot turn it off. Grok 4.3 reasons by default, so on simple tasks watch the reasoning count in your own usage.
Which Grok model for which task
- Chat answers: Grok 4.3 at 0.45¢ beats Grok 4.7 at 0.96¢. Both are above the 0.24¢ median, so for high-volume chat a cheaper model is better.
- Blog posts: Grok 4.3 at 1.00¢ against 2.3¢ for Grok 4.7. DeepSeek V4.1 Flash at 0.37¢ is cheaper still.
- Ticket classification: Grok 4.3 at 1.0¢ against 2.2¢. DeepSeek V4.1 Flash at 0.63¢ is cheaper than both.
- Agents: Grok 4.3 at 73¢ against $1.70 for Grok 4.7. DeepSeek V4.1 Flash at 16¢ is far below both.
Grok 4.7 scores 1444 on LMArena Text against 1442 for Grok 4.3, a small gap. I would pick Grok 4.7 only if you have measured a quality difference on your own work. Otherwise Grok 4.3 is the Grok model to use, and for these four tasks a cheaper model is usually the better buy.
To test your own prompts, use the calculator. For how I measured tokens and costs, see the methodology. Task pages: chatbot answer, blog post, ticket classification and 50-step agent. Model pages: Grok 4.7 and Grok 4.3.
FAQ
Is Grok 4.7 more expensive than Grok 4.3 per task?
Yes, on all four tasks I compared. Its list price is higher ($2.00 input and $6.00 output per 1M tokens, against $1.25 and $2.50), and the token counts were identical in my runs.
Does reasoning always raise the cost?
Reasoning tokens are billed as output, so more reasoning means a higher bill for the same visible answer. Qwen3.8 Flash used 16,000 reasoning tokens on the blog post and returned no visible output, but still cost 0.75¢.
What is the cheapest model for a 50-step agent here?
Of the four models compared, DeepSeek V4.1 Flash at 16¢ per task, then Gemini 3.5 Flash Lite at 24¢, Grok 4.3 at 73¢ and Grok 4.7 at $1.70. These are modeled figures, so check them against your own workload.