Grok beta
xAI introduced Grok as a conversational assistant with access to current information through X.
- Conversational assistant
- Real-time X context
- Early beta access
← AI Tools directory
General AssistantResearch dossier
Grok
A general and developer assistant focused on real-time web/X context, agentic coding, long-running work and visual creation.
Grok is the consumer product and model brand. Grok 4.6 is available through developer/partner surfaces and the API; consumer access, tools and model selection vary across grok.com, X and apps.
First releasedNovember 4, 2023
CompanySpaceXAI
Best forCoding, agent workflows, real-time web/X context and productivity integrations
PricingFree and paid access varies by product/plan; API pricing is model-specific
What Grok is today
Grok began as an assistant integrated with X and quickly expanded from conversational access into multimodal models, APIs, search tools and coding agents.
Grok 4.6 emphasizes sustained, multi-step agent work and stronger first-pass visual or interactive projects rather than only benchmark question answering.
Reading note
Product features and underlying models are documented separately. Roadmap items include only provider-announced work; rumors are excluded.
Major versions and what changed
Selected milestones that materially changed the product.
Grok-1.5
Reasoning and coding improved while context expanded to 128K tokens.
- MATH rose from 23.9% to 50.6%
- HumanEval rose from 63.2% to 74.1%
- 16x previous context length
- Long-context retrieval evaluation
Grok 4 and 4.1 Fast
Grok moved into deeper reasoning, agent tool use and a 2M-context fast tool-calling variant.
- Web and X search tools
- Code execution
- File retrieval
- MCP and long-horizon tool calling
Grok 4.6
The current flagship improves longer agent trajectories, coding and ambitious visual or interactive work.
- More self-testing on long tasks
- Improved visual first passes
- 500K context on Vertex AI
- Configurable low-to-xhigh reasoning
Grok 4.5 High vs Grok 4.6 High
The comparison uses provider-published capabilities and named evaluations, not a generic marketing score.
| Dimension | Previous | Current | Measured difference |
|---|---|---|---|
| CursorBench 3.2 | 66.7% | 69.9% | +3.2 points (about 4.8% relative) |
| DeepSWE 1.1 | 54.0% | 65.9% | +11.9 points (about 22% relative) |
| Terminal-Bench 3.0 | 15.7% | 26.0% | +10.3 points (about 65.6% relative) |
| Long-running behavior | Strong agent and coding baseline | More persistence, self-testing and refinement | Better fit for multi-step projects according to xAI testing |
The gain ranges from about 1.05x to 1.66x on selected agent benchmarks.
Those ratios are calculated from xAI's published Grok 4.5 and 4.6 scores. They are benchmark-specific and should not be generalized to every task.
DeepSWE relative score
65.9% versus 54.0% for Grok 4.5 High.
Check evidenceTerminal-Bench relative score
26.0% versus 15.7%; this is the largest relative gain among the selected published comparisons.
Check evidence4.6 context window
The Vertex AI release lists a 500,000-token context window and configurable reasoning effort.
Check evidenceHow the important features work
What happens behind the button, where it helps and what to verify.
Real-time web and X search
Product and API tool access variesHow it works: The agent can issue web or X searches, inspect results and incorporate recent information into a response.
Best use: Breaking news, public sentiment and current research
Watch: Fast-moving social posts can be wrong or manipulated; verify primary evidence.
Agent Tools API
API pricing and tool-call charges applyHow it works: Server-managed web search, X search, code execution, file retrieval and MCP tools can be selected and called over multiple turns.
Best use: Support agents, finance workflows and research automation
Grok Build
Availability and included usage varyHow it works: Uses the model as a coding agent to research, structure, implement and iterate on software projects.
Best use: Prototypes, codebase work and interactive applications
Watch: Run tests and review security-sensitive changes before deployment.
Grok Imagine
Web, mobile and API availability variesHow it works: Generates or transforms images and video; current video models can create synchronized motion, sound and speech from prompts or starting images.
Best use: Creative ideation and short-form media
LIMITS
What the headline features do not guarantee
- xAI's reported benchmarks use specific reasoning settings and harnesses; results may not match consumer chat behavior.
- Real-time X access increases freshness but also exposes the system to low-quality or adversarial content.
- High-effort agent tasks can be slower and need careful supervision when tools can take actions.
Announced roadmap, without speculation
Status reflects the provider's latest public wording. “No public date” means there is no verified launch schedule.
Rolling out
Broader Grok 4.6 distribution
xAI launched 4.6 across its API and partners, including Vertex AI shortly after the initial release.
Official announcementAnnounced
Longer, self-verifying agents
xAI identifies sustained work, self-testing and ambitious interactive projects as the focus of the 4.6 line.
Official announcementNo public date
Next numbered Grok
No verified launch date or feature specification for a Grok 5 generation is public.
Primary sources
Provider announcements, help documentation and system information.
- 1Announcing Grok-1.5
SpaceXAI · Mar 28, 2024
- 2Grok 4.1 Fast and Agent Tools API
SpaceXAI · Nov 19, 2025
- 3Introducing Grok 4.6
SpaceXAI · Aug 12, 2026
Last editorial verification: 2026-08-24. Prices, access limits, preview status and model routing can change after this date.