TokenMix Research Lab · 2026-04-24

Grok API Key: How to Get Access + Pricing 2026

Grok API Key: How to Get Access + Pricing 2026

Getting a Grok API key in 2026 requires an xAI developer account — the process takes 5 minutes and gives you access to Grok 3, Grok 4 Fast, Grok 4.1 Fast (Reasoning and Non-Reasoning), and Grok 4.20 Beta. Pricing: $3 input / 5 output per MTok on the flagship Grok 4 models, $0.50/$2.50 on Grok 4 Fast, with Grok 4.1 Fast Reasoning at $2.50/ 2.50. Free tier: limited to 10 requests/min on Grok 4 Fast Non-Reasoning. This guide covers the signup walkthrough, how to verify your key works, pricing comparisons vs GPT-5.4 and Claude Opus 4.7, and the SpaceX-xAI IPO context that's driving rate-limit tightening. TokenMix.ai exposes all Grok variants via one OpenAI-compatible endpoint — useful if you want Grok + multi-provider fallback.

Table of Contents

Confirmed vs Speculation

Claim Status Source
Grok API access via x.ai/api Confirmed xAI developer portal
Grok 4 $3/ 5 per MTok Confirmed Pricing docs
Grok 4 Fast $0.50/$2.50 Confirmed
Free tier 10 req/min Fast Non-Reasoning Confirmed Rate limit page
OpenAI-compatible endpoint Confirmed API reference
xAI acquired by SpaceX Feb 2026 Confirmed SpaceX-xAI merger
Grok 5 release date announced No — speculation
API reliability Variable — 2 outages in April 2026 status.x.ai

xAI Signup in 5 Minutes

Step 1 — Create xAI account:

Step 2 — Add billing:

Step 3 — Generate API key:

Step 4 — Test:

curl https://api.x.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_XAI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-4-fast-non-reasoning",
    "messages": [{"role":"user","content":"Hello Grok"}]
  }'

Success = JSON response with choices[0].message.content.

Grok Model Pricing Tiers

Model Input $/MTok Output $/MTok Context Best for
grok-3 $0.30 .50 131K Budget general
grok-3-mini $0.15 $0.60 131K Cheap chat
grok-4 $3.00 5.00 1M Flagship general
grok-4-fast-non-reasoning $0.50 $2.50 2M Latency-critical
grok-4-fast-reasoning $2.00 0.00 2M Fast reasoning
grok-4.1-fast-non-reasoning $0.50 $2.50 2M Current fast
grok-4.1-fast-reasoning $2.50 2.50 2M Current reasoning
grok-4.20 Beta $3.00 5.00 2M 4-agent multi

Key tiers:

First API Call: curl + Python

Python using OpenAI SDK (Grok is OpenAI-compatible):

from openai import OpenAI

client = OpenAI(
    api_key="your_xai_key",
    base_url="https://api.x.ai/v1"
)

response = client.chat.completions.create(
    model="grok-4.1-fast-reasoning",
    messages=[{"role": "user", "content": "Explain dark matter in 3 sentences."}]
)

print(response.choices[0].message.content)

Or via TokenMix.ai for multi-provider routing:

client = OpenAI(
    api_key="your_tokenmix_key",
    base_url="https://api.tokenmix.ai/v1"
)
# Now call model="xai/grok-4.1-fast-reasoning"

vs GPT-5.4 and Claude Opus 4.7

Model Input $/MTok Output $/MTok SWE-Bench Verified Non-halluc. rate
Grok 4 $3.00 5.00 ~70% 80%
Grok 4.20 Beta $3.00 5.00 ~70% 83% (4-agent)
GPT-5.4 $2.50 5.00 58.7% 76%
Claude Opus 4.7 $5.00 $25.00 87.6% 82%
Gemini 3.1 Pro $2.00 2.00 80.6% 75%

Grok 4.20's 4-agent architecture is the differentiator for reasoning-sensitive queries. For pure coding, Claude Opus 4.7 wins. For general chat at low cost, GPT-5.4 or Grok 3 Mini.

Common Errors When Setting Up

Error: 401 Unauthorized → API key wrong or missing Bearer prefix in header

Error: 403 Rate limit exceeded → Free tier only gives 10 req/min. Upgrade billing tier or switch to paid model.

Error: Model not found: grok-4 → Use exact ID grok-4 or grok-4-0709 — case-sensitive. Newer models may require explicit version suffix.

Error: Service unavailable (503) → xAI outage, check status.x.ai. Common during IPO-related demand spikes. Use multi-provider fallback.

Error: Context length exceeded → Some variants (Grok 3) cap at 131K. Switch to Grok 4 or Grok 4 Fast for 1M-2M context.

FAQ

Does Grok have a free API tier?

Yes — grok-4-fast-non-reasoning offers 10 requests/minute free. Enough for dev prototyping. For production, paid tier required.

What's the difference between Grok 4 and Grok 4.20?

Grok 4 is single-model flagship. Grok 4.20 Beta adds 4-agent parallel architecture (Grok + Harper + Benjamin + Lucas) with 83% non-hallucination rate. Grok 4.20 is 3-4× slower due to cross-verification. Use Grok 4 for latency-critical, 4.20 for accuracy-critical. See Grok 4.20 review.

Is Grok reliable for production?

Moderate reliability. Two multi-hour outages in April 2026 post-SpaceX-merger. For mission-critical paths, always implement fallback routing through TokenMix.ai or similar gateway.

What rate limits should I expect?

Tier 1 (new account): 10 req/min free model, 60 req/min paid. Tier 4 (after $400+ spend): 200-400 req/min. Custom enterprise tiers negotiate higher. See xAI rate limits.

Can I use Grok for coding?

Yes, but SWE-Bench Verified ~70% lags Claude Opus 4.7 (87.6%) and GLM-5.1. For production coding, Claude or GLM. For general chat where code questions occasionally come up, Grok is adequate.

Does Grok support image input?

Grok 4 and 4.20 have vision capabilities. Send images via standard OpenAI-compatible image message format. Quality is solid but below Claude Opus 4.7's 3.75MP visual acuity.

How does Grok handle politically sensitive content?

Less restricted than Claude or GPT on political topics — one of xAI's brand differentiators. For production with safety requirements, add your own guardrails — Grok's default behavior is more permissive.

Sources

By TokenMix Research Lab · Updated 2026-04-24