xAI Grok
by xAI (a SpaceX subsidiary)
Frontier-tier reasoning at roughly a third of rival output pricing, with native X and web search
Grok is xAI's model family, spanning grok-4.5 for frontier reasoning, grok-4.3 and the 4.20 line at million-token context, and grok-build-0.1 for coding. It couples top-tier measured intelligence with output pricing well below comparable frontier models, plus native Web Search and X Search tools no competitor can license.
Grok is the model family from xAI, founded by Elon Musk on 9 March 2023 in Palo Alto and acquired by SpaceX as a wholly-owned subsidiary on 2 February 2026 at a $250 billion valuation. The current API line-up spans grok-4.5, described by xAI as its most intelligent and fastest model, with a 500,000-token context window at $2.00 per million input tokens, $0.30 cached and $6.00 output; grok-4.3 and the grok-4.20-0309 reasoning, non-reasoning and multi-agent variants at a 1,000,000-token window and $1.25/$0.20/$2.50; and grok-build-0.1, a coding-focused model with a 256,000-token window at $1.00/$0.20/$2.00. Pricing rises above those rates for prompts of 200,000 tokens or more, so the large windows carry a premium. Knowledge cutoff across the line is 1 February 2026 — the most recent of any major provider. Beyond text, xAI ships Grok Imagine for image input and video generation, speech-to-speech, speech-to-text and text-to-speech APIs, and native Web Search and X Search tools that give the models live access to X data no competitor can license. Independent measurement by Artificial Analysis puts grok-4.5 at high reasoning effort at an intelligence index of 54, ranking 13th of 185 models against a median of 32, but also records output at 56.1 tokens per second — 105th of 185 and below the 69 median — with time-to-first-token of 9.16 seconds. On data handling xAI states it never trains on API inputs or outputs without explicit permission, retains requests for 30 days for auditing and then deletes them, and offers optional zero data retention that is signalled per request through an x-zero-data-retention response header, while advising most customers against enabling it because it disables several features. It publishes SOC 2 Type 2 and GDPR compliance, with a BAA available on request through a questionnaire process and further certification detail gated behind an NDA at its trust centre.
Cost-conscious platform teams running high-volume reasoning workloads that can absorb slower token throughput in exchange for roughly a third of comparable frontier output pricing.
Top-15 measured intelligence at $2 input and $6 output per million tokens, with the most recent knowledge cutoff of any major provider and live X data no rival can offer.
At a Glance
- Category
- AI Models & APIs
- Pricing
- Usage-based
- Target Market
- CTOs, CIOs, Enterprise Developers, Data Scientists, Platform Engineers
- Deployment
- API-based, Cloud-only
- Founded
- 2023
- Headquarters
- Palo Alto, United States
- Team Size
- 500+
Key Features
- ✓grok-4.5 frontier model
500,000-token context at $2.00 input and $6.00 output per million, rated 13th of 185 models by Artificial Analysis.
- ✓Million-token context on grok-4.3 and the 4.20 line
A 1,000,000-token window at $1.25 input per million, among the cheapest large-context options available anywhere.
- ✓Native Web Search and X Search tools
Live retrieval including X platform data, a real-time source no competing model vendor is able to license.
- ✓Full modality coverage
Grok Imagine for image input and video generation, plus speech-to-speech, speech-to-text and text-to-speech from one vendor.
- ✓grok-build-0.1 coding model
A coding-specialised model with a 256,000-token window at $1.00 input and $2.00 output per million tokens.
- ✓Per-request zero data retention signalling
An x-zero-data-retention response header lets applications programmatically confirm whether ZDR was active for that call.
- ✓1 February 2026 knowledge cutoff
The most recent cutoff among major providers, reducing how much recent context has to be supplied through retrieval.
Capabilities
Use Cases
- •Real-time social and market monitoring
Use native X Search to reason over live platform discussion that no other model vendor can access directly.
- •Cost-optimised high-volume reasoning
Move bulk reasoning traffic to grok-4.5 where $6 per million output substantially undercuts comparable frontier models.
- •Million-token document and codebase analysis
Run whole archives through grok-4.3's 1M window at $1.25 per million input for large-scale review work.
- •Voice-enabled enterprise assistants
Combine speech-to-text, reasoning and text-to-speech from one vendor rather than stitching three providers together.
- •Agentic coding pipelines
Route software-engineering tasks to grok-build-0.1 at $1.00 input per million with a 256,000-token window.
Ideal For
Best For
- ✓High-volume reasoning workloads where output token cost dominates the bill and $6 per million beats comparable frontier pricing
- ✓Applications needing live social and web signal, via native X Search and Web Search tools that only xAI can provide
- ✓Long-context analysis on grok-4.3 or the 4.20 line at a million-token window and $1.25 per million input
- ✓Products needing text, vision, image and video generation plus speech-to-speech, speech-to-text and text-to-speech from a single vendor
- ✓Teams that want a per-call reasoning-effort dial and the most recent knowledge cutoff available, at 1 February 2026
Not Ideal For
- ✗Latency-sensitive interactive products — Artificial Analysis measured grok-4.5 at 56.1 tokens per second against a 69 median, with 9.16 seconds to first token
- ✗Organisations where brand and moderation risk are procurement gates: independent reporting ties slow enterprise and government adoption to regulatory investigations over explicit AI-generated imagery and weak moderation safeguards
- ✗Buyers requiring ISO 27001 — xAI publishes SOC 2 Type 2 and GDPR, and further certification detail sits behind an NDA at its trust centre
- ✗Teams that need self-serve HIPAA coverage, since xAI's BAA requires a questionnaire and manual review rather than a click-through
- ✗Anyone pinning models for long-term stability, given concurrent grok-4.20, 4.3, 4.5 and build-0.1 lines and a rapid version cadence
Integrations
Deployment
Market Analysis
Pros
- ✓Artificial Analysis rates grok-4.5 at intelligence index 54, 13th of 185 models against a median of 32, at $2 input and $6 output per million
- ✓Knowledge cutoff of 1 February 2026 is the most recent of any major provider, reducing dependence on retrieval for recent facts
- ✓Native Web Search and X Search provide live social signal that competitors structurally cannot match
- ✓Context windows of 500K on grok-4.5 and 1M across the 4.3 and 4.20 lines, at low input pricing
- ✓Complete modality coverage — text, vision, image, video, and speech in three directions — under one vendor
- ✓Clear data posture: no training on API data without explicit permission, 30-day audit retention then deletion, and optional ZDR confirmed per request via response header
Cons
- ✗Independent reporting describes slow enterprise and government adoption, with Grok limited to pilots and low-priority tasks such as document drafting and paid adoption below ChatGPT and Claude
- ✗The same reporting attributes that to reputational damage from controversial content generation, including allegations of explicit AI-generated imagery and weak moderation safeguards, plus regulatory investigations in multiple regions
- ✗Artificial Analysis measured output at 56.1 tokens per second, 105th of 185 and below the 69 median, with 9.16 seconds to first token
- ✗Prompts of 200,000 tokens or more are priced above the headline rates, so the large context windows are a paid upgrade rather than an included feature
- ✗No public evidence of ISO 27001 certification, and detailed compliance evidence sits behind an NDA at the trust centre
- ✗Zero data retention is enterprise-only and xAI itself advises most customers against enabling it because it disables several features
- ✗Rapid, overlapping version cadence across 4.20, 4.3, 4.5 and build-0.1 makes model pinning and re-qualification a continuing cost
Pricing
grok-4.5
From $2.00 per 1M input tokens
- ✓$2.00 input, $0.30 cached, $6.00 output per 1M tokens under 200K
- ✓500,000-token context window
- ✓Reasoning-effort control per call
- ✓Web Search and X Search tools
grok-4.3 and grok-4.20-0309
From $1.25 per 1M input tokens
- ✓$1.25 input, $0.20 cached, $2.50 output per 1M tokens under 200K
- ✓1,000,000-token context window
- ✓Reasoning and non-reasoning variants
- ✓Multi-agent variant available
grok-build-0.1
From $1.00 per 1M input tokens
- ✓$1.00 input, $0.20 cached, $2.00 output per 1M tokens under 200K
- ✓256,000-token context window
- ✓Coding-focused model
Metered per token with no published seat pricing for API access: grok-4.5 at $2.00 input, $0.30 cached and $6.00 output per million, grok-4.3 and the 4.20 line at $1.25/$0.20/$2.50, and grok-build-0.1 at $1.00/$0.20/$2.00. The catch buyers should model is the 200,000-token threshold — prompts at or above it are billed above these headline rates, so the 500K and 1M windows cost more than the sticker price implies. Zero data retention is enterprise-gated and disables several features.
Security & Compliance
Connect
Sources
This page was written from 5 sources, 5 on domains other than x.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
TwelveLabs
Video intelligence API that makes every hour of enterprise footage searchable, analyzable and agent-ready.
Engram
A learned memory layer that makes AI actually know your organization — at up to 100x fewer tokens.
Mistral OCR 4
Structure-aware document AI that returns bounding boxes, typed blocks, and per-word confidence scores.
Hugging Face
The AI community building the future.