Grok 4.20 vs Grok 4.3 Thinking

Same price, different habits

Grok 4.20 and Grok 4.3 are two xAI models released about seven weeks apart in spring 2026, and both are in our free chat. They even cost the same through the API. What separates them is how they work: in our chat, Grok 4.20 answers immediately without a reasoning step, while Grok 4.3 Thinking reasons quietly for a moment before replying. That one difference decides which to use for a given question.

Grok 4.20 vs Grok 4.3 in numbers

Prices: OpenRouter list price, Oct 2026.

Where each came from

Grok 4.20 first appeared as a beta in grok.com in February 2026, where it worked as a team of four cooperating agents. When it reached the API in March, xAI offered it in three forms: a reasoning version, a non-reasoning version and a multi-agent version that runs 4 or 16 agents. Our chat uses the plain model with reasoning switched off, which makes it the quickest model we offer.

Grok 4.3 arrived at the end of April as a single model with adjustable reasoning, from none up to extra high. It was xAI's newest general model until Grok 4.5 arrived in July. We run it with low reasoning effort and hide the reasoning text, so you see only the final answer.

Speed vs care

Because it does not stop to think, Grok 4.20 usually starts writing within about a second. That makes it pleasant for chatting, quick rewrites, translations, short summaries, simple facts and brainstorming. Its weakness is the flip side: on tricky questions it can commit to the first plausible answer.

Grok 4.3 Thinking takes a few seconds longer because it works through the question first. That pays off when there is a trap or several steps: maths word problems, logic puzzles, planning with constraints, comparing options with numbers, or debugging. If you have ever had a quick model confidently give a wrong total, this is the model to switch to.

A quick test you can run

Ask both models the same question with a twist, such as a schedule that crosses midnight or a discount applied twice. Grok 4.20 will often get it right, but when it slips, Grok 4.3 Thinking usually catches the detail. You can try this in our chat: ask Grok 4.20, then switch the model and ask again.

For long pasted text, both have large windows, though our chat keeps only the recent part of a very long conversation. Starting a fresh chat for a new topic helps either model.

Which model for which job

None of these is a strict rule. If Grok 4.20's answer looks wrong, ask the same thing in Grok 4.3 Thinking before you rely on it. The reverse also works: if Grok 4.3 gives a long, careful answer to a simple request, Grok 4.20 will usually give you the short version faster. Since both are free here within the same daily allowance of 30 messages, switching costs you nothing but one message.

For developers

On xAI's API the two models are addressed differently. Grok 4.20 is published as two dated variants, one with reasoning and one without (the March 9 snapshot, hence "0309" in the ids), plus a multi-agent option. Grok 4.3 is one model, grok-4.3, and you set how hard it thinks with a reasoning effort parameter. On OpenRouter the ids are x-ai/grok-4.20 and x-ai/grok-4.3, and reasoning is switched on, off or tuned in the request. Because prices match, the choice comes down to latency and accuracy, not budget. Keep in mind that hidden reasoning tokens are billed as output, so Grok 4.3 at higher effort costs more per answer even at the same list price. Code samples are in our Grok API guide.

Choose Grok 4.20 if…

• you want the fastest replies for everyday writing and questions;
• the task is simple: rewording, translating, summarizing, listing ideas;
• you are chatting back and forth and waiting would break the flow;
• you are a developer who needs the larger context window offered on OpenRouter.

Chat with Grok 4.20

Choose Grok 4.3 if…

• the question has several steps or a hidden catch;
• you need calculations, logic or a plan that has to add up;
• a wrong answer would cost you more than a few seconds of waiting;
• you want a documented knowledge cut-off (December 2025).

Chat with Grok 4.3 Thinking. For the hardest tasks, also see Grok 4.7 vs Grok 4.3 and the model pages for Grok 4.20 and Grok 4.3.

FAQ

Which is newer, Grok 4.20 or Grok 4.3?

Grok 4.3. It came out on April 30, 2026, about seven weeks after Grok 4.20 reached the API.

Why is Grok 4.20 faster?

In our chat it runs with reasoning switched off, so it starts the answer straight away. Grok 4.3 Thinking reasons briefly first.

Do they cost the same?

Yes, per token: $1.25 in and $2.50 out per million on OpenRouter. Grok 4.3 may use more output tokens because of its reasoning. In our chat both are free within the daily allowance.

Can I switch between them in one conversation?

Yes. Open the model menu at the top of our chat and pick the other one; the choice is saved for that conversation.