xAI

Grok 4.3

Provider

xAI

Status

Current

Context

1,000,000 tok

Price

$1.25 / $2.5 /MTok

Knowledge

2025-12

Grok 4.3 is xAI’s current flagship model, released in beta on grok.com and the SuperGrok apps on 17 April 2026 and via the API on 30 April 2026. It is a reasoning model with a 1M-token context window, and the first xAI model with native video input — it can take video clips, transcribe speech, segment speakers, track objects and reason about motion. Its standout commercial trait is price: at $1.25 / $2.50 per million tokens it is one of the cheapest frontier-adjacent models available, well below Claude Opus 4.8 ($5/$25) and GPT-5.5 ($5/$30).

On capability, Grok 4.3 is strong on science and hard reasoning — about 90.1% on GPQA Diamond and a 53.2 Artificial Analysis Intelligence Index — but xAI concedes it trails Claude Opus 4.7 on agentic coding (SWE-bench Pro) by double-digit points, and xAI has not published a full SWE-bench table for it (Artificial Analysis, llm-stats). The model is also inseparable from its company context: xAI is now a subsidiary of SpaceX, and the Grok apps’ image tools remain under regulatory scrutiny after an early-2026 deepfake scandal (see the xAI provider page).

Quick specs

TRY GROK 4.3 →

What’s new in Grok 4.3

Grok 4.3 is an incremental flagship over Grok 4.2, with three changes that matter in practice (codersera, Artificial Analysis).

Native video input

Grok 4.3 is the first xAI model to accept video natively, taking mp4, mov and webm clips up to five minutes at 1080p. It can transcribe speech, segment speakers, track objects across frames and reason about motion — useful for analysing recordings, demos and clips without a separate transcription step.

Document generation

The model can produce structured documents directly — PDFs, spreadsheets and slides — rather than only returning text or code, narrowing the gap with assistant features in ChatGPT and Gemini.

Aggressive pricing

At $1.25 input and $2.50 output per million tokens, Grok 4.3 is priced well below the other flagships, with an additional $0.20/MTok cached-input rate. The main caveat: requests over 200,000 input tokens are billed at double the standard rate, so very long-context calls cost $2.50/$5.00.

Benchmark performance

xAI did not publish a full first-party benchmark table for Grok 4.3, so the most reliable figures come from Artificial Analysis, with τ²-Bench from xAI’s materials. Where a number is not available from a credible source, it is marked accordingly rather than estimated.

The read-through is consistent: Grok 4.3 is genuinely strong on graduate-level science (GPQA Diamond ~90%) and hard reasoning, and competitive on tool-use benchmarks like τ²-Bench. But its Artificial Analysis Intelligence Index of 53.2 places it mid-pack among current flagships rather than at the very top, and xAI’s own framing is that it trails Claude Opus 4.7 on agentic coding (SWE-bench Pro) by double-digit percentage points (llm-stats). For coding specifically, Claude Opus 4.8 (69.2% SWE-bench Pro) and GPT-5.5 are clearly ahead. On throughput, Grok 4.3 outputs around 171 tokens/second but, as a reasoning model, has a slow time-to-first-token (~12 seconds) (Artificial Analysis).

Pricing breakdown

Grok 4.3’s pricing is its sharpest competitive edge.

Cost comparison with contemporaries

For high-volume reasoning where top-tier coding isn’t the priority, Grok 4.3’s token economics are hard to beat among hosted frontier-class models.

How to access Grok 4.3

Via API

Grok 4.3 is generally available with no waitlist on xAI’s OpenAI-compatible API as grok-4.3, so existing OpenAI integrations can switch with minimal changes.

from openai import OpenAI
client = OpenAI(base_url="https://api.x.ai/v1", api_key="XAI_API_KEY")

resp = client.chat.completions.create(
    model="grok-4.3",
    messages=[{"role": "user", "content": "Your prompt here"}],
)
print(resp.choices[0].message.content)

Via the Grok apps

Access depends on subscription tier. The free tier uses Grok 4-class models rather than the 4.3 flagship; paid tiers add Grok 4.3 plus Grok Imagine media generation and (on SuperGrok) voice companions.

Grok is also available inside the X app and in Tesla vehicles. See the Grok app page for the full consumer breakdown.

How Grok 4.3 compares

Grok 4.3’s mid-2026 contemporaries are Claude Opus 4.8, GPT-5.5 and Google’s Gemini 3.5 Pro, plus its own predecessor Grok 4.2.

vs Claude Opus 4.8

Claude Opus 4.8 is clearly ahead on agentic coding (69.2% SWE-bench Pro, 88.6% SWE-bench Verified) and professional knowledge work, and is the better choice when output quality on engineering or long autonomous runs is the priority. Grok 4.3’s counter is price — roughly a quarter of Opus 4.8’s output cost — and native video input. Choose Opus 4.8 for coding and reliability; choose Grok 4.3 for cheap, high-volume reasoning and video understanding.

vs GPT-5.5

A clean, same-harness head-to-head against GPT-5.5 is data not available from primary sources, since xAI did not publish matching SWE-bench numbers. Directionally, GPT-5.5 leads on agentic and terminal coding, while Grok 4.3 competes on science reasoning (GPQA Diamond ~90%) and undercuts GPT-5.5 heavily on price ($1.25/$2.50 vs $5/$30).

vs Gemini 3.5 Pro

Google’s Gemini 3.5 Pro is the newer Google flagship and a strong all-rounder with a large context window; a verified same-harness comparison with Grok 4.3 is data not available, so we won’t invent one. Both compete on value, with Gemini strong on multimodal and Workspace integration and Grok strong on real-time X data. See the Google hub for the current Gemini lineup.

vs Grok 4.2

Grok 4.3 supersedes Grok 4.2 (February 2026), adding native video input, document generation and a refreshed price. Grok 4 Heavy remains the separate multi-agent variant for the hardest reasoning, available on SuperGrok Heavy.

The practical consensus

Grok 4.3 is a capable, very cheap reasoning model that punches above its price on science and analysis, but it is not the model to reach for on frontier-level coding, where Claude and GPT lead. See our best AI models and best AI for coding rankings for where it sits across the wider field.

Known limitations

No published SWE-bench. xAI has not released SWE-bench Verified or Pro for Grok 4.3, and concedes it trails Claude Opus 4.7 on agentic coding — so its coding standing is, at best, mid-pack and partly unverifiable.

Mid-pack intelligence index. A 53.2 Artificial Analysis Intelligence Index (≈38th) puts it below the top flagships on the most-cited independent composite.

Slow first token. As a reasoning model, time-to-first-token is around 12 seconds, which is noticeable in interactive use even though sustained output is fast.

Long-context cost cliff. Requests over 200,000 input tokens are billed at double the standard rate.

Company and safety baggage. Grok’s app-side image tools remain under regulatory scrutiny after the early-2026 deepfake scandal, and xAI publishes less safety documentation than its peers — relevant for risk-averse and enterprise use. See the xAI provider page.

Version history

xAI is training Grok 5 on the Colossus 2 supercomputer, with a reported ~10-trillion-parameter target and a projected arrival in the second or third quarter of 2026.

Frequently asked questions

When was Grok 4.3 released?

Grok 4.3 launched in beta on grok.com and the SuperGrok apps on 17 April 2026, with public API access following on 30 April 2026. It is xAI’s current flagship model.

How much does Grok 4.3 cost?

Via the xAI API, Grok 4.3 is $1.25 per million input tokens and $2.50 per million output tokens, with cached input at $0.20/MTok. Requests over 200,000 input tokens are billed at double the standard rate. In the apps, it is included with SuperGrok Lite ($10/mo) and above.

Is Grok 4.3 good at coding?

It is capable but not class-leading. xAI has not published SWE-bench scores for Grok 4.3 and concedes it trails Claude Opus 4.7 on agentic coding (SWE-bench Pro) by double-digit points. For top coding performance, Claude Opus 4.8 and GPT-5.5 are stronger.

What can Grok 4.3 do that earlier Grok models could not?

Grok 4.3 is the first xAI model with native video input — it can take video clips and transcribe speech, segment speakers, track objects and reason about motion — and it can generate documents such as PDFs, spreadsheets and slides.

What is Grok 4.3’s context window?

One million tokens, with input across text, images and video. Note that requests over 200,000 input tokens are billed at twice the standard rate.

Is Grok 4.3 better than Claude or ChatGPT?

It depends on the job. Grok 4.3 is cheaper and strong on science reasoning (about 90% on GPQA Diamond) and real-time X data, but Claude Opus 4.8 and GPT-5.5 lead on agentic coding and carry stronger safety and enterprise track records. See best AI for coding for the head-to-head.

Last verified 18 June 2026. Benchmark figures are from Artificial Analysis and xAI; xAI has not published SWE-bench scores for Grok 4.3, which are marked data not available rather than estimated. Pricing and availability current as of the publication date and subject to change.