Grok 4.3

What is disclosed

Grok 4.3 went live on the xAI API on 5 May 2026, according to an xAI post that day.¹¹1xAI, post on X announcing that Grok 4.3 is live on the xAI API, 5 May 2026, 16:39 UTC. https://x.com/xai/status/2051703217697010103. The time comes from the post's metadata on X's syndication service and matches the post ID: (2051703217697010103 >> 22) + 1288834974657 ms is 5 May 2026, 16:39 UTC. The post said it topped Artificial Analysis leaderboards in agentic tool calling and instruction following. It also claimed first place on Vals AI tests in fields such as case law and corporate finance.¹¹

The model page describes it as fast, with strong tool calling and instruction following.²²2xAI, Grok 4.3 model page, docs.x.ai, read 29 September 2026. https://docs.x.ai/developers/models/grok-4.3 Nothing about its size or design is public.

It holds 1,000,000 tokens of context.²² No output limit is stated. It reads images as well as text and writes only text. Each image can be up to 20 MiB, in JPEG or PNG, and a request can carry any number of them.³³3xAI, Models overview, docs.x.ai, last updated 21 September 2026. https://docs.x.ai/developers/models

Reasoning is optional. Effort none switches it off, and the default is low.²² The model supports function calling and structured outputs.²²

What is new

Its prices and its 1M-token window are those of the Grok 4.20 models, live since 10 March 2026.⁴⁴4xAI, API release notes: Grok 4.20 on 10 March, Grok 4.5 on 8 July, Grok 4.6 on 12 August and Grok 4.7 on 21 September 2026. https://docs.x.ai/developers/release-notes ⁵⁵5xAI, API pricing, docs.x.ai, read 29 September 2026: grok-4.3 and the three grok-4.20 models at $1.25 input, $0.20 cached and $2.50 output, with 1M tokens of context; from 200K prompt tokens, $2.50, $0.40 and $5.00. https://docs.x.ai/developers/pricing. Arithmetic: 2.50 / 1.25 = 0.40 / 0.20 = 5.00 / 2.50 = 2.

On 15 May 2026 xAI retired a group of older models, grok-4-0709 and grok-3 among them.⁶⁶6xAI, Grok model retirement on 15 May 2026, migration guide, last updated 2 September 2026. https://docs.x.ai/developers/migration/may-15-retirement. Arithmetic: of the eight retired slugs, grok-code-fast-1 goes to Grok Build 0.1 and grok-imagine-image-pro to an image model, so 8 minus 2 = 6 go to Grok 4.3. Grok 4 Fast and Grok 4.1 Fast went too, each in a reasoning and a non-reasoning version.⁶⁶ Their slugs still resolve. Six of them now send requests to Grok 4.3, billed at Grok 4.3 prices.⁶⁶

A redirected reasoning model runs at low effort, a non-reasoning one at none.⁶⁶ xAI tells customers to choose a replacement explicitly, to avoid unexpected cost increases and to control the effort they pay for.⁶⁶

The retirement guide calls Grok 4.3 "the fastest, most intelligent model" xAI had built. Three newer models have followed since, the latest being Grok 4.7 on 21 September 2026.⁶⁶⁴⁴

Using it

The model id is grok-4.3, and the alias grok-4.3-latest follows the newest version.²²³³ Input costs $1.25 per million tokens and output $2.50.⁵⁵ Cached input is $0.20 per million. From 200K prompt tokens on, the whole request is billed at double these rates.⁵⁵

Unlike Grok 4.7, it takes Batch API requests.²²⁷⁷7xAI, Grok 4.7 model page, docs.x.ai, read 29 September 2026: Batch API not supported, 500,000 tokens of context, $6 per million output tokens. https://docs.x.ai/developers/models/grok-4.7. Arithmetic: 1,000,000 / 500,000 = 2; 2.50 / 6 = 0.42. It runs in us-east-1, with default limits of 37 requests per second and 10,000,000 tokens per minute.²²

Beside Grok 4.7, it has twice the context and costs less than half as much per output token.⁷⁷