Grok 4.3

xAI frontier model with context caching and live web grounding. Cost control leans on cache hits and alias routing rather than async discounts.

$1.25
Input / 1M tokens
$2.50
Output / 1M tokens
$0.20
Cache read · −84%
1M
Context window

At a glance.

ProviderxAI
Familygrok-4
Released2026-04
LicenseProprietary
Context window1M tokens
Max output30K tokens
Modalitiestext, image, pdf
Tool callingYes
Reasoning modeYes
Cachingcontext
Batch discountNo batch tier

What reuse looks like here.

Grok 4.3 · agent trafficper request
Total input100%
Candidate reuse58%
Realized reuse44%
Capture rate76%
280ms
Warm TTFT · −56% vs cold
130
Output tokens / sec
Reuse economics

What you actually pay once caching works.

At a typical 55% prefix reuse, a million input tokens on Grok 4.3 effectively costs $0.67 instead of $1.25 - blending to roughly $1.13 with a 25% output share. There is no batch tier, so cost control here leans on caching and routing.

Estimate it for your workload
Best for
generalrealtime-knowledge

Routes through these aliases:

Same OpenAI client, this model.

python
from openai import OpenAI

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

r = client.responses.create(
    model="grok-4-3",          # or an alias like auto.best
    input="Draft a fix for the failing test.",
)
print(r.usage.input_tokens_cached)   # confirm reuse

Grok 4.3, answered.

How much does Grok 4.3 cost?

Grok 4.3 is $1.25 per million input tokens and $2.50 per million output tokens through Zumik. Cache reads are $0.20 per million, a 84% discount on input.

What is Grok 4.3's context window?

Grok 4.3 supports a 1M-token context window with up to 30K output tokens.

Does Grok 4.3 support prompt caching?

Yes. xAI uses Context caching caching. In the Zumik corpus, Grok 4.3 shows a median cache capture of 76% on agent workloads.

Which Zumik aliases route to Grok 4.3?

Grok 4.3 is a candidate for the auto.best alias, selected when it wins under current routing policy.

Run Grok 4.3 with reuse measured.

Point an OpenAI client at Zumik and see exactly how much of this model's input you are reusing.