As of November 9, 2026, xAI describes Grok 4.7 as its most capable model for coding and knowledge work, and its model overview recommends it for general work. Its API model ID is grok-4.7. The API catalog also lists Grok 4.20 Reasoning and a Grok 4.20 Multi-Agent entry. Model numbers alone do not establish release order, so this guide identifies Grok 4.7 as the documented general-purpose flagship—not as a claim that no other listed model is newer. (xAI announcement; xAI model overview)
Which Grok API model should you evaluate?
| Workload to evaluate | API listing | What the documentation supports |
|---|---|---|
| General coding and knowledge work | Grok 4.7 (grok-4.7) |
xAI describes it as its most capable model for coding and knowledge work. The developer guide lists a 500,000-token context window and support for function calling, web search, X search, and code execution. (1, 3) |
| A reasoning-focused alternative | Grok 4.20 Reasoning (grok-4.20-0309-reasoning) |
The model listing gives a 1,000,000-token context window. The available documentation does not establish its image-input or function-calling details. (4) |
| Parallel-agent research | Grok 4.20 Multi-Agent (grok-4.20-multi-agent-0309) |
The listing describes multiple agents collaborating on deep-research tasks. The available evidence does not establish its context limit or current release/support status, so confirm those details before making it a production dependency. (5) |
Start with Grok 4.7 for general API work, then evaluate another listed model if its documented purpose matches your workload. The available excerpts do not establish that the alternatives share Grok 4.7’s endpoint, request format, or all of its tools; check each model’s own documentation before integrating it.
Grok API pricing
The xAI pricing documentation states that prices are in USD. The model listings show these input and output rates per million tokens: (6)
| API model | Input | Output |
|---|---|---|
| Grok 4.7 | $2.00 / 1M tokens | $6.00 / 1M tokens |
| Grok 4.20 Reasoning | $1.25 / 1M tokens | $2.50 / 1M tokens |
| Grok 4.20 Multi-Agent | Not established by the available pricing evidence | Not established by the available pricing evidence |
The available evidence does not establish cached-input rates or a context-length threshold that changes pricing, so those are omitted. Check current model-specific pricing before estimating costs, especially for Multi-Agent. These are API token rates, not consumer-app subscription prices. (3–6)
Example estimate: At the listed Grok 4.7 rates, 100,000 input tokens and 20,000 output tokens would total $0.32 USD in token charges: (100,000 × $2.00 ÷ 1,000,000) + (20,000 × $6.00 ÷ 1,000,000). This estimate excludes any unverified cache, tool, or context-dependent charges.
Make a production API request
The Grok 4.7 developer guide shows a Responses API request to https://api.x.ai/v1/responses, using bearer-token authentication. Set your xAI API key in the XAI_API_KEY environment variable before making the request. (3)
curl https://api.x.ai/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $XAI_API_KEY" \
-d '{
"model": "grok-4.7",
"input": "Find and fix the bug, then explain it: function median(a){a.sort();return a[a.length/2]}"
}'
This example is specifically for grok-4.7. Do not assume that changing the model ID alone makes the same request compatible with either Grok 4.20 listing; consult that model’s current API instructions first.
How to decide for your workload
- Select prompts representing real tasks, including common and difficult cases.
- Compare only models available to your API account, keeping inputs and tool conditions consistent where possible.
- Score task success and output quality against explicit requirements; also record latency and API cost.
- Include failure checks, such as whether a model admits uncertainty rather than inventing a source.
- Route by use case only after reviewing your results.
For example, compare Grok 4.7 with Grok 4.20 Reasoning on representative long-document work. Include the Multi-Agent listing only if parallel research is relevant, and first confirm its current status, limits, and price. Published positioning can help choose candidates, but does not establish a winner for your workload.
What remains to verify
The available evidence does not settle Multi-Agent’s current release/support status, context limit, or price, cached-input charges, or whether the Grok 4.20 entries use the same request interface as Grok 4.7. Verify those details in current xAI documentation before committing to a production configuration.