Grok 4.3
Grok 4.3 is SpaceXAI's budget workhorse: a 1-million-token model for $1.25 in and $2.50 out per million tokens. It reached SuperGrok Heavy subscribers as a beta on 17 April 2026 and the xAI API on 30 April 2026. When SpaceXAI retired Grok 4, Grok 4 Fast and Grok 4.1 Fast on 15 May 2026, it sent all their traffic to Grok 4.3.
Artificial Analysis scored it 53 on its Intelligence Index at launch (on the older, pre-v4.3 scale), 4 points above Grok 4.20, while cutting input price by 37.5% and output price by 58%. It is no longer SpaceXAI's smartest model; Grok 4.7 is. But it is still the cheapest current Grok, has twice 4.7's context window, and is the only current one that supports the cheaper Batch API.
Grok 4.3 is the Grok to pick when cost and context matter more than top-end intelligence.
Why it still earns a place:
- Price. $1.25 / $2.50 is well under the $2 / $6 of Grok 4.5 to 4.7, and cached input is just $0.20.
- Long context. 1 million tokens, double Grok 4.7's 500,000.
- Batch support. It works with the Batch API for bulk jobs; Grok 4.5, 4.6 and 4.7 do not.
- Tool calling. Artificial Analysis measured 98% on tau2-Bench Telecom, a test of following a customer-service script with tools, and 81% on IFBench for instruction following.
The catches:
- It talks a lot. Artificial Analysis found it used about 44% more output tokens than Grok 4.20, which eats part of the price cut.
- Weaker at hard coding. SpaceXAI now points coders to Grok 4.7, and Artificial Analysis said Grok 4.3 trailed GPT-5.5 by 276 Elo on its GDPval office-work test.
Choose it for chatbots, data extraction, long-document questions and batch jobs. Skip it for complex coding agents; use Grok 4.7 instead.
Score breakdown
Best for
- Customer-service bots and tool-calling agents on a budget
- Question answering over very long documents
- Overnight batch processing
- Replacing retired Grok 4 and Grok 4.1 Fast integrations
What we like
- Cheapest current Grok: $1.25 / $2.50 per 1M tokens, $0.20 cached
- 1M-token context, twice Grok 4.7's
- Batch API support for bulk, non-urgent jobs
- Strong tool use: 98% on tau2-Bench Telecom (Artificial Analysis)
- Reasoning can be switched off entirely for fast, cheap replies
Watch out for
- Clearly behind Grok 4.7 and rival frontier models on hard coding
- Uses about 44% more output tokens than Grok 4.20
- Default low reasoning effort can surprise teams expecting deep thinking
- Silent redirects from retired models can change costs for old integrations
Specs at a glance
| Developer | SpaceXAI (formerly xAI) |
|---|---|
| API name | grok-4.3 (alias grok-4.3-latest) |
| Release dates | Beta for SuperGrok Heavy: 17 April 2026; API: 30 April 2026; wider app rollout: early May 2026 |
| Context window | 1,000,000 tokens |
| Input / output | Text and image in; text out |
| Reasoning effort | none, low (default), medium, high, xhigh |
| API pricing (under 200K prompt) | $1.25 input, $0.20 cached input, $2.50 output per 1M tokens |
| API pricing (200K+ prompt) | $2.50 input, $0.40 cached input, $5.00 output per 1M tokens |
| Batch API | Supported |
| Rate limit | 37 requests per second |
| Replaces | grok-4-0709, grok-4-fast, grok-4-1-fast and grok-3 (redirected since 15 May 2026) |
| Predecessor / successor | Grok 4.20 / Grok 4.5 |
Benchmarks
Benchmarks are standard tests. Vendor-run results are marked as such; independent results are preferred where they exist.
| Benchmark | Score | Source | Note |
|---|---|---|---|
| Artificial Analysis Intelligence Index (Apr 2026) | 53 | Artificial Analysis | On the older (pre-v4.3) scale; 4 points above Grok 4.20; just above Claude Sonnet 4.6. Later index versions are rescaled. |
| GDPval-AA (Elo) | 1,500 | Artificial Analysis | Up 321 from Grok 4.20; about 276 below GPT-5.5 (xhigh) |
| tau2-Bench Telecom | 98% | Artificial Analysis | Up 5 points from Grok 4.20 |
| IFBench | 81% | Artificial Analysis | |
| Cost to run the AA Intelligence Index | $395 | Artificial Analysis | About 20% less than Grok 4.20 |
Pricing
| Plan / tier | Price | Notes |
|---|---|---|
| API input | $1.25 per 1M tokens | $2.50 when the prompt is 200K tokens or more |
| API cached input | $0.20 per 1M tokens | $0.40 above 200K |
| API output | $2.50 per 1M tokens | $5.00 above 200K |
| Batch API | Discounted batch rates | Shown on the xAI model page |
| Grok app | Free and SuperGrok Lite tiers | Lower plans have used Grok 4.3 with limits, per third-party plan trackers |
What Grok 4.3 changed
Grok 4.3 followed Grok 4.20 and moved in the opposite direction on price. SpaceXAI focused on faster reasoning, better instruction following and agent tool use, and cut the price sharply.
| Grok 4.20 (launch) | Grok 4.3 | |
|---|---|---|
| Input price | $2.00 | $1.25 |
| Output price | $6.00 | $2.50 |
| AA Intelligence Index (older, pre-v4.3 scale) | 49 | 53 |
| GDPval-AA | about 1,179 | 1,500 |
Grok 4.20 figures are derived from Artificial Analysis's stated gains. In the app, Grok 4.3 also added the ability to create downloadable PDFs, spreadsheets and slide decks from a chat, according to launch reviews.
The 15 May 2026 retirement
On 15 May 2026 SpaceXAI retired eight older API models. Requests to these names still work, but they now run on Grok 4.3:
| Old model | Now routed to |
|---|---|
| grok-4-0709 (Grok 4) | Grok 4.3, low reasoning |
| grok-4-1-fast-reasoning | Grok 4.3, low reasoning |
| grok-4-1-fast-non-reasoning | Grok 4.3, no reasoning |
| grok-4-fast (both modes) | Grok 4.3, low or no reasoning |
| grok-3 | Grok 4.3, no reasoning |
| grok-code-fast-1 | grok-build-0.1 |
All redirected calls are billed at Grok 4.3 prices. For users of Grok 4.1 Fast, which cost $0.20 / $0.50, that is a large price rise.
Safety
SpaceXAI did not publish a detailed safety report for Grok 4.3 that we could find. Artificial Analysis found its non-hallucination rate on AA-Omniscience fell 8 points from Grok 4.20, which means it was more likely to give a wrong answer than admit it did not know. Add fact-checking steps for anything that matters. For the company's wider record see our SpaceXAI page.
- Set reasoning effort yourself. The default is
low; usenonefor simple lookups andmediumorhighfor multi-step problems. - If you were redirected from Grok 4.1 Fast, compare your bill before and after 15 May 2026. You may be paying about 6x more for input.
- Use the Batch API for jobs that can wait. Grok 4.3 supports it; the $2 / $6 models do not.
- Keep prompts below 200K tokens where you can. The 1M window is there, but everything above 200K is billed at double rates.
Jargon explained
- Batch API
- A way to send many requests at once and get the results later, usually at a discount.
- tau2-Bench
- A test where an AI plays a support agent and must use tools and follow policy correctly to solve a customer's problem.
- Hallucination
- When an AI states something false with confidence, instead of saying it does not know.
Alternatives to consider
Frequently asked questions
When was Grok 4.3 released?
SuperGrok Heavy subscribers got a beta on 17 April 2026. It reached the xAI API on 30 April 2026 and rolled out more widely in the Grok app in early May.
How much does Grok 4.3 cost?
$1.25 per million input tokens and $2.50 per million output tokens, with cached input at $0.20. Prompts of 200,000 tokens or more cost double.
Why am I being charged Grok 4.3 prices for Grok 4?
SpaceXAI retired grok-4-0709 and several other models on 15 May 2026. Calls to those names now run on Grok 4.3 and are billed at its rates.
Is Grok 4.3 better than Grok 4.7?
No. Grok 4.7 is smarter and much better at coding. Grok 4.3 wins on price, context length (1M vs 500K tokens) and Batch API support.
Sources
Every fact on this page comes from public information. Vendor figures are labelled as vendor claims.
- Grok 4.3 model page (SpaceXAI)
- May 15, 2026 Model Retirement (SpaceXAI)
- xAI launches Grok 4.3 with improved agentic performance and lower pricing (Artificial Analysis)
- xAI rolls out Grok 4.3 beta for SuperGrok Heavy subscribers (PiunikaWeb)
- Grok 4.3 Review: What's New in xAI's Latest Model (DEV Community)
- Grok 4.3 scores 53 on Artificial Analysis Intelligence Index (OfficeChai)
- Grok Pricing for SuperGrok subscription plans (Suprmind)