Grok Model Updates 2026: Current Versions, API Access & What Changed

As of June 2026, xAI's Grok model lineup looks nothing like it did six months ago. Grok 4.3 replaced the original Grok 4 as the recommended API model. Eight legacy model IDs — including grok-3, grok-4-0709, and several fast-reasoning variants — were retired on May 15, 2026, and now silently redirect to Grok 4.3. If you haven't audited your API calls recently, you may already be running on a different model than you think.

This is a developer reference guide. It covers every current Grok model, their model IDs, pricing, context windows, rate limits, and what changed from the legacy lineup.

Current Grok Models (June 2026)

xAI currently offers six production API models. Grok 4.3 is the recommended default for all general-purpose workloads.

Prices sourced from xAI Models docs, retrieved June 6, 2026.

Grok 4.3 — The Default

Grok 4.3 launched on April 30, 2026 and is xAI's recommended model for all API users. It replaced Grok 4 (grok-4-0709) as the default alias target.

Key specs:

• Context window: 1,000,000 tokens (1M)
• Output speed: ~207 tokens/second
• Input modalities: text, image (≤20 MiB), video (new in 4.3)
• Reasoning: always-on, cannot be toggled off
• Knowledge cutoff: November 2024
• Rate limits: 1,800 requests/min, 10M tokens/min
• Regions: us-east-1, us-west-2, eu-west-1

Grok 4.3 introduced three capabilities absent in the 4.20 generation: native video input (up to 5 minutes per video), document generation (PDF/XLSX/PPTX output), and improved agentic tool calling.

Source: xAI Models — Grok 4.3, retrieved June 6, 2026.

Grok 4.20 — Three Specialized Variants

Grok 4.20 shipped in March 2026 in three variants, all priced at $1.25/$2.50 per 1M tokens:

• Reasoning (grok-4.20-0309-reasoning): 1M context, chain-of-thought optimized.
• Non-reasoning (grok-4.20-0309-non-reasoning): 1M context, standard completions without reasoning overhead. Better for latency-sensitive workloads.
• Multi-agent (grok-4.20-multi-agent-0309): 2M context window. Designed for multi-agent orchestration where multiple agents share a long context.

The 2M multi-agent variant has the largest context window in xAI's lineup. For document analysis pipelines or complex agent loops that need extreme context, this is the model to pick.

Note: logprobs and top_logprobs are not supported on Grok 4.20 and newer. These fields are silently ignored if set. Source: xAI docs.

Grok 4.1 Fast — Best Price-to-Performance

At $0.20/$0.50 per 1M tokens with a 2M context window, Grok 4.1 Fast is the cost-optimized option in xAI's lineup. With prompt caching, the cached input rate drops to $0.05/M — effectively negligible at scale.

Grok 4.1 Fast delivers near-flagship quality at roughly 1/6th the input cost and 1/5th the output cost of Grok 4.3. For chatbot, RAG, and most agent workloads, this is the practical default if budget matters more than maximum reasoning depth.

Source: mem0.ai Grok API Pricing, retrieved June 6, 2026.

Grok Code Fast 1 — The New Default for Agentic Coding

Grok Code Fast 1 (grok-code-fast-1) launched in early June 2026 and is now xAI's recommended model for agentic coding and fast tool-calling workloads. It replaces Grok Build 0.1 as the model behind the Grok Build CLI and serves at 100+ tokens/second — fast enough for interactive terminal workflows.

Key specs:

• Context window: 256,000 tokens (256K)
• Pricing: $1.00/M input, $2.00/M output
• Output speed: 100+ tokens/second
• Strengths: Code generation, debugging, function calling, structured outputs
• Reasoning: supported via the reasoning parameter

Outside of coding, it's also a strong option for general-purpose agentic and tool-calling use cases where throughput matters more than maximum intelligence. At $1/$2 per 1M tokens, it's a fraction of the cost of Grok 4.3.

Source: xAI Release Notes — June 2026; Releasebot — xAI Updates, retrieved June 8, 2026.

Grok Build 0.1 — Legacy Coding Model

Grok Build 0.1 was xAI's first dedicated coding model, available since May 2026. It runs via the grok CLI (installable with curl -fsSL https://x.ai/cli/install.sh | bash) or through the API. Priced at $1.00/$2.00 per 1M tokens with a 256K context window — the same pricing tier as its successor, Grok Code Fast 1.

Grok Build 0.1 is now a legacy alias. For new integrations, use grok-code-fast-1 instead.

Source: xAI Release Notes — May 2026.

Legacy Model Retirement (May 15, 2026)

On May 15, 2026, xAI retired eight legacy model IDs. Requests to these IDs are now automatically redirected to Grok 4.3:

If your integration uses any of these model IDs, it still works — but you're now running on Grok 4.3 whether you intended to or not. The output quality, pricing, and behavior characteristics may differ from what you originally tested against.

The grok-4.3 model page lists all aliases that now resolve to it, including grok-latest, grok-3-latest, grok-4-latest, and many others.

Source: xAI Release Notes; Basenor — Grok Gets Major Upgrades, retrieved June 6, 2026.

API Access & Setup

The Grok API is OpenAI SDK compatible. To use it with the openai Python package:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_XAI_API_KEY",
    base_url="https://api.x.ai/v1",
)

response = client.chat.completions.create(
    model="grok-4.3",
    messages=[{"role": "user", "content": "Explain quantum computing in one paragraph."}],
)

You can also use the xAI SDK directly:

import os
from xai_sdk import Client
from xai_sdk.chat import user

client = Client(api_key=os.getenv("XAI_API_KEY"))
chat = client.chat.create(model="grok-4.3")
chat.append(user("Explain quantum computing"))
response = chat.sample()
print(response.content)

API keys are managed at console.x.ai. The endpoint is https://api.x.ai/v1.

Source: x.ai/api, retrieved June 6, 2026.

Multi-Provider Routing

If you're already using TheRouter to route across OpenAI-compatible providers, adding Grok is straightforward — it uses the same /v1/chat/completions endpoint format. This lets you set up fallback routing between Grok 4.3, GPT-4.1, Claude Sonnet, and others with a single integration.

For guidance on routing strategies, see our LLM API Fallback Routing post (coming soon) and the OpenAI-Compatible API Providers reference.

Pricing Comparison: Grok vs Competitors

How does Grok 4.3 stack up against other frontier models? All prices per 1M tokens, as of May 2026.

Grok 4.3 is 37.5% cheaper on input and 68.75% cheaper on output than GPT-4.1. Against Claude Sonnet 4.6, it's 58% cheaper on input and 83% cheaper on output. Grok 4.1 Fast is in a different category entirely at $0.20/$0.50.

The tradeoff: xAI's developer ecosystem is smaller than OpenAI's or Anthropic's, with fewer third-party integrations and less mature tooling. For teams already invested in the OpenAI ecosystem, Grok's OpenAI-compatible endpoint makes it easy to add as a fallback or cost-optimized alternative without rewriting application code.

Source: mem0.ai Grok API Pricing, May 2026.

Server-Side Tools & Additional Costs

Beyond token costs, Grok's API offers server-side tools billed per invocation:

File attachments are the most expensive tool at $10/1K calls. For cost-sensitive applications, design workflows to minimize redundant tool calls and cache results where possible.

Source: xAI docs; mem0.ai.

Image & Video Generation

Voice API

Source: xAI docs — Models.

Model Alias System

xAI uses a three-tier alias system:

• <modelname> → aliases to the latest stable version. Recommended for most users.
• <modelname>-latest → aliases to the very latest version. Use this if you want cutting-edge features immediately.
• <modelname>-<date> → pinned to a specific release. Use this for workflows that demand reproducibility.

For example, grok-4.3 is the stable alias and will continue to resolve to the latest 4.3.x release. grok-4.3-latest gives you the absolute newest version. grok-4.20-0309-reasoning is pinned to the March 9 release.

Source: xAI docs — Model Aliases.

What's Coming: Grok 5 and Beyond

xAI is actively training Grok 5 on Colossus 2, its next-generation 1-gigawatt supercomputer cluster. Reports indicate intermediate checkpoints (Grok 4.2, 4.4, 4.5) with a target release window of Q2–Q3 2026 for the full Grok 5 variant.

This section is speculative — xAI has not published an official Grok 5 release date or pricing. We'll update this reference when official details are available.

Source: Basenor — Grok Gets Major Upgrades, May 2026.

FAQ

What is the current Grok model for API use?

Grok 4.3 (grok-4.3) is xAI's recommended default as of June 2026. It's the most intelligent general-purpose model in their lineup, with a 1M context window and $1.25/$2.50 pricing.

Is the Grok API OpenAI-compatible?

Yes. The Grok API uses the same /v1/chat/completions endpoint format as OpenAI. You can use the openai Python/Node.js SDK by changing base_url to https://api.x.ai/v1 and swapping in your xAI API key.

What happened to Grok 3 and the original Grok 4?

Both were retired on May 15, 2026. Requests to grok-3, grok-4, and other legacy model IDs are now automatically redirected to grok-4.3. Your code still works, but you're running on Grok 4.3 without realizing it.

How much does Grok 4.3 cost?

$1.25 per 1M input tokens and $2.50 per 1M output tokens. Cached input costs $0.20/M. This makes it cheaper than GPT-4.1, Claude Sonnet 4.6, and Gemini 2.5 Pro on a per-token basis.

What is the cheapest Grok model?

Grok 4.1 Fast at $0.20/M input and $0.50/M output. With prompt caching, the cached input rate drops to $0.05/M. It has a 2M token context window.

Does Grok support logprobs?

Not on Grok 4.20 and newer. The logprobs and top_logprobs fields are silently ignored. If you need logprobs for scoring or classification, you'll need a different model.

What is Grok Code Fast 1?

Grok Code Fast 1 (grok-code-fast-1) is xAI's latest coding model, launched in June 2026. It serves at 100+ tokens/second at $1.00/$2.00 per 1M tokens with a 256K context window. It's optimized for agentic coding, function calling, and structured outputs. It now powers the Grok Build CLI, replacing the earlier grok-build-0.1 model.

What is Grok Build?

Grok Build is xAI's terminal-native AI coding agent, launched in May 2026. It was originally powered by grok-build-0.1 but now runs on grok-code-fast-1 ($1.00/$2.00 per 1M tokens, 256K context, 100+ tok/s). It's currently in early access, primarily available to SuperGrok Heavy subscribers ($300/month).

How do I get free Grok API credits?

xAI has offered up to $150/month in free API credits through its data-sharing program (enable "Share API Inputs for Model Training" in the xAI console). Verify current availability — the program has been subject to changes.