Grok 4.3 is xAI's flagship API model as of April 30, 2026. It's a reasoning model with a 1-million-token context window, native video input, real-time access to X (Twitter) data, and a knowledge cutoff of December 2025. xAI completed the full API rollout on April 30 alongside an approximately 40% price cut on input tokens. Pricing is $1.25 per million input tokens and $2.50 per million output tokens. The model targets agentic workflows, instruction-following, multi-step investigations, and applications requiring high factual accuracy. It's available via the xAI API and Oracle Cloud Infrastructure Generative AI.

How is Grok 4.3 different from Grok 4.20?

Five major changes: (1) Context window grew to 1M tokens (up from 256K on Grok 4.20). (2) Input pricing cut approximately 40% to $1.25 per million tokens. (3) Native video input support — first time for the Grok family. (4) Improved architecture for advanced logic, math, scientific analysis, and multi-step investigations. (5) Stronger agentic tool use and instruction following. xAI's official model docs recommend all API callers migrate from Grok 4.20 to Grok 4.3.

How much does Grok 4.3 cost in May 2026?

Grok 4.3 costs $1.25 per million input tokens and $2.50 per million output tokens via the xAI API as of April 30, 2026. The output token price has no hard limit (unlimited output tokens per call). The pricing represents roughly a 40% input-token cut from Grok 4.20, making Grok 4.3 the lowest-cost frontier model from a US-based closed-weights provider as of May 2026. Only DeepSeek V4 variants (open weights, Chinese provider) come in cheaper at the same capability tier.

Where can I use Grok 4.3?

Grok 4.3 is available through three main channels in May 2026: (1) The xAI API directly at api.x.ai with the model name `grok-4.3`. (2) Oracle Cloud Infrastructure Generative AI as `xai.grok-4.3` — became available one day after public release (around May 8). (3) The Grok consumer apps on X/Twitter for end users. Grok 4.3 is not currently available on Amazon Bedrock or Google Cloud Vertex AI.

Quick Answer

What Is Grok 4.3? xAI's 1M-Context Model (May 2026)

Published: May 11, 2026

What Is Grok 4.3? xAI’s 1M-Context Model (May 2026)

Grok 4.3 is xAI’s flagship API model as of April 30, 2026 — a reasoning model with a 1-million-token context window, native video input, real-time X data access, and aggressive pricing that resets the cost model for agentic apps.

Last verified: May 11, 2026

Quick facts

Property	Value
Vendor	xAI
Full API rollout	April 30, 2026
OCI availability	~May 8, 2026
Context window	1,000,000 tokens
Knowledge cutoff	December 2025
Input price	$1.25 per 1M tokens
Output price	$2.50 per 1M tokens
Output token limit	None (unlimited)
Native video input	Yes
Real-time X data	Yes
Open weights	No
Model id (xAI)	`grok-4.3`
Model id (OCI)	`xai.grok-4.3`

What’s new vs Grok 4.20

Grok 4.3 is positioned as a major architectural and pricing upgrade over Grok 4.20.

1. 1M-token context (up from 256K). Long-context refactors, multi-document analysis, full-codebase audits — all become tractable on a single call.

2. ~40% input price cut. Input dropped to $1.25 per million tokens. Combined with the larger context, this materially changes the unit economics of long-context agentic apps.

3. Native video input. First time for the Grok family. The model can take video frames directly as input — useful for video understanding agents, multimodal investigations, and content analysis pipelines.

4. Better reasoning architecture. Improved performance on advanced logic, math, scientific analysis, and multi-step investigations. xAI describes Grok 4.3 as a reasoning model suited for “accuracy-critical tasks.”

5. Stronger agentic tool use. Instruction-following and tool calling are tighter — important for production agent loops.

Benchmarks

Grok 4.3 doesn’t lead the SWE-bench leaderboards (Claude Opus 4.7 and GPT-5.5 do), but it’s competitive across the board:

Artificial Analysis Coding Index: 41.0 (better than 89% of compared models)
Long-context retention past 128K: Among the top performers (alongside GPT-5.5)
Reasoning benchmarks: Strong on math, logic, multi-step analysis
Outperforms GPT-5.1 on private legal and financial benchmarks (per third-party evals)

The model isn’t trying to win SWE-bench — it’s trying to be a credible third frontier option at a fraction of the price.

Unique capabilities

Real-time X (Twitter) data. Grok 4.3 can pull live posts, trends, and replies from X as part of a query. For news monitoring, social listening, current-events agents, this is genuinely differentiated — no other frontier model has this.

Native video input. Pass a video file and Grok 4.3 processes it directly. Useful for video understanding, surveillance analytics, content moderation pipelines, video QA.

Unlimited output tokens. No hard cap per response — useful for long-form generation, full-document refactors, full-codebase explanations.

1M-token context at the lowest frontier price. Combine 1M context with $1.25 input pricing and the cost of “load this whole repo and answer questions” drops by an order of magnitude vs Opus 4.7.

Pricing in context

Model	Input/1M	Output/1M
Grok 4.3	$1.25	$2.50
DeepSeek V4-Pro	$1.74	$3.48
DeepSeek V4-Flash	$0.14	$0.28
Gemini 3.1 Pro	mid	mid
GPT-5.5	mid	mid
Claude Opus 4.7	$5	$25

Among closed-weights frontier models, Grok 4.3 is the cheapest. DeepSeek V4 variants undercut on raw price but are open-weights and Chinese-provider — different procurement story.

Where to use it

1. xAI API directly — api.x.ai, model grok-4.3. Best for direct integration.

2. Oracle Cloud Infrastructure Generative AI — model xai.grok-4.3. Available one day after public release (~May 8). Best for OCI customers and enterprise procurement.

3. Grok consumer apps on X — bundled in the Grok product on x.com and the X mobile apps. End-user facing.

Not currently available on Amazon Bedrock or Google Cloud Vertex AI.

When to pick Grok 4.3

Pick Grok 4.3 when:

Real-time X data is part of the workflow.
Native video input matters.
1M-token context at the lowest frontier price is the deciding factor.
You want a credible third option besides Anthropic and OpenAI.
Cost matters more than the absolute top SWE-bench score.

Don’t pick Grok 4.3 when:

You need open weights or self-hosting (use DeepSeek V4-Pro).
You need the top SWE-bench Verified score (use Claude Opus 4.7).
Terminal-Bench performance is critical (use GPT-5.5).
You need MCP-Atlas top performance (use Claude Opus 4.7).

What to watch next

Grok 5 — rumored for later 2026.
Bedrock / Vertex AI availability — would meaningfully expand procurement options.
Independent benchmarks maturing for the 1M-context performance claim.
Pricing wars — DeepSeek V4 ran a 75%-off promo through May 5; xAI may respond.

Last verified: May 11, 2026 — sources: xAI Grok 4.3 docs, Oracle Cloud Grok 4.3 docs, RoboRhythms release coverage, WinZheng analysis, DataStudios characteristics breakdown, ApiYi release notes.