Grok 4.6 launched on August 12, 2026. The clean launch-day verdict is that this is a meaningful agent and coding upgrade over Grok 4.5, not a context-window expansion or a price cut. The xAI API model page lists a 500,000-token context window; Cursor lists 256K for its separate Grok 4.6 integration. xAI reported stronger results across all ten benchmarks in its launch table. Its August 12 launch announcement named the API, Grok Build, Cursor and gateway partners including OpenRouter, Vercel and Cloudflare. xAI later announced GitHub Copilot (August 14), Gemini Enterprise Agent Platform (August 21) and Microsoft Foundry (August 26). These are dated announcements, not guarantees of current plan or account access.
The August 12 launch API rate was $2 per million input tokens and $6 per million output tokens, but that shorthand is incomplete. It applies to short-context requests. Once a prompt enters xAI’s long-context band at 200,000 tokens, the rate doubles to $4 per million input tokens and $12 per million output tokens—and xAI says the higher rate applies to all tokens in that request. That caveat is the difference between a useful price comparison and a misleading one.
| Grok 4.6 fact | Source-backed detail |
|---|---|
| Release date | August 12, 2026 |
| API model | grok-4.6 |
| xAI API context window | 500,000 tokens |
| Price below 200K prompt tokens | $2 input / $0.50 cached input / $6 output per 1M tokens |
| Long-context price at the 200K band | $4 input / $1 cached input / $12 output per 1M tokens |
| Inputs and outputs | Text and image input; text output |
| Reasoning levels | Low, medium, high (default), and xhigh |
| Dated access announcements | August 12 launch: xAI API, Grok Build, Cursor, OpenRouter, Vercel and Cloudflare; later xAI announcements: GitHub Copilot (August 14), Gemini Enterprise Agent Platform (August 21), Microsoft Foundry (August 26). Check each provider’s current listing and terms. |
What Grok 4.6 Is
xAI describes Grok 4.6 as its frontier model for coding, agentic tasks, and knowledge work. The company says the update focuses on long-running agents and more ambitious interactive and visual projects: researching unfamiliar domains, working across codebases, using tools, building first versions, and continuing through multiple rounds of feedback.
The official training description is specific about the direction, though not about model size. Grok 4.6 received a longer supplemental training run than Grok 4.5, using curated model-generated reasoning and technical data, engineering data, a revised optimizer, regenerated supervised-fine-tuning trajectories, and reinforcement learning across knowledge work, coding, kernel optimization, web development, computer-aided design, and other agent environments. xAI does not publish a parameter count in the launch post or API documentation.
That framing matters. This is not a new consumer chatbot tier with a bigger memory number. It is an API-and-agent release built to stay useful over longer trajectories. The current xAI model page lists function calling, structured outputs, and reasoning for Grok 4.6. The page’s generic web, X, and code links do not establish those tools as Grok 4.6 capabilities.
Grok 4.6 Benchmarks: The Full Official Table
xAI’s benchmark table shows Grok 4.6 improving over Grok 4.5 on every reported evaluation. It matches GPT-5.6 Sol Max at 61 on the Artificial Analysis Intelligence Index, leads the comparison on GDPVal-AA v2, AA-Briefcase, and Harvey LAB, and remains behind the best comparison score on CursorBench, DeepSWE, FrontierCode, APEX-Agents, Terminal-Bench, and APEX-SWE.
| Evaluation | Grok 4.6 High | Grok 4.5 High | GPT-5.6 Sol Max | Fable 5 Max |
|---|---|---|---|---|
| AA Intelligence Index | 61 | 56 | 61 | 62 |
| GDPVal-AA v2 | 1753 | 1526 | 1728 | 1741 |
| CursorBench v3.2 | 69.9% | 66.7% | 67.2% | 70.5% |
| DeepSWE v1.1 | 65.9% | 54.0% | 73.0% | 70.0% |
| FrontierCode v1.1 Extended | 61.3% | 56.6% | 60.6% | 63.6% |
| APEX-Agents | 57.5% | 47.1% | 56.7% | 59.2% |
| Terminal-Bench v3.0 | 26.0% | 15.7% | 34.6% | 34.1% |
| APEX-SWE | 56.4% | 53.6% | Not listed | 58.8% |
| AA-Briefcase | 1577 | 1313 | 1502 | 1574 |
| Harvey LAB (Vals) | 15.8% | 12.9% | 2.5% | 11.3% |
The improvement over Grok 4.5 is substantial in several places: DeepSWE rises by 11.9 percentage points, Terminal-Bench by 10.3 points, APEX-Agents by 10.4 points, and the AA Intelligence Index by five points. The more important competitive read is mixed. Grok 4.6 is at or above GPT-5.6 Sol Max on seven of the nine rows where xAI lists a GPT score, but it trails GPT clearly on DeepSWE and Terminal-Bench. Fable 5 Max remains ahead on the headline Intelligence Index and six task-specific rows.
These are vendor-reported launch results, not a neutral Kingy test. xAI says third-party scores are the best of self-reported or publicly available figures, which means harnesses, reasoning settings, tool access, and test conditions may differ. The right conclusion is that Grok 4.6 has moved forward and belongs in serious agent evaluations—not that one launch chart establishes a universal winner.
Grok 4.6 Pricing: The 200K Caveat
Current global API prices checked October 11, 2026: xAI’s pricing table lists $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens below 200,000 prompt tokens. At or above that threshold, the long-context rates are $4, $1, and $12 per million respectively; xAI bills all tokens in the request at the long-context rate. The US regional endpoint costs 1.1 times global token rates, a 10% premium. These global rates match the figures in the August 12 launch table; confirm the pricing and endpoint before purchase. See the official xAI pricing table.
| Request band | Input | Cached input | Output |
|---|---|---|---|
| Short context, below 200K prompt tokens | $2 / 1M | $0.50 / 1M | $6 / 1M |
| Long context, xAI’s ≥200K band | $4 / 1M | $1 / 1M | $12 / 1M |
The billing rule is more important than the doubled numbers: once the prompt reaches the long-context threshold, xAI bills all tokens in that request at the long-context rates. It is not a surcharge applied only to the portion above 200K.
A simple example shows the effect. A request with 100,000 input tokens and 10,000 output tokens costs about $0.26 at standard rates. A request with 250,000 input tokens and 10,000 output tokens costs about $1.12 at long-context rates. That second prompt is 2.5 times larger, but the bill is more than four times higher because the entire request moves into the doubled tier.
xAI also announces a fast variant at twice the price. The launch material does not publish a distinct model slug for that variant on the Grok 4.6 overview page, so check the live console before hard-coding a name. Do not confuse the fast premium with the 200K long-context tier; they are separate pricing caveats, and the public launch page does not explain whether or how they combine.
xAI’s October 11 pricing table lists cached input below the standard input rate. For cache-key headers and context compaction, use the current API guide and verify compatibility with Grok 4.6; the model and pricing captures cited here do not establish those implementation details.
What the 500K Context Window Really Means
The xAI API documents a 500,000-token context window for Grok 4.6. Cursor currently lists 256K for its Grok 4.6 integration, so the usable limit depends on the provider. That is enough for large repositories, long research packets, multi-document analysis, or extensive agent histories—but it is not an instruction to fill every request. The long-context price switch makes context management part of the product decision.
The context window also did not grow from Grok 4.5, which the historical comparison lists at 500K; that earlier figure was not confirmed by the current source check. The upgrade is the model’s reported ability to use context and sustain work, not a larger headline limit.
How to Access Grok 4.6
1. Use Grok 4.6 in the xAI API
The current xAI model page lists the API model as grok-4.6 and provides an API playground. Use the linked model documentation to confirm request details and account terms before integrating.
For a current request example, use the xAI model-specific API documentation.
For production integrations, pin the explicit grok-4.6 string rather than assuming an older latest alias has moved. Monitor the usage object and your xAI console because team access, rate limits, regional availability, and live pricing can vary.
2. Select Grok 4.6 in Cursor
Cursor currently lists Grok 4.6 in its Models pool for individual and team plans. Choose it from the model picker for an agent or chat. If it is not visible immediately, refresh or update Cursor and check the Models section in Cursor settings. The first-week double-included-usage promotion has ended, so check your current plan allowance rather than budgeting from the launch offer.
3. Use Grok Build
In its August 12 launch post, xAI said Grok 4.6 was available in Grok Build and offered double included usage for the first week. That launch promotion has elapsed. This source check does not establish current Grok Build plan or allowance terms.
4. Select Grok 4.6 in GitHub Copilot
xAI announced Grok 4.6 availability in GitHub Copilot on August 14. Its announcement described the model picker in VS Code, Copilot CLI and cloud agents; some business and enterprise administrators may need to enable the model in Copilot settings. Check the current Copilot model picker and plan allowance.
5. Use a Model Gateway
xAI’s August 12 launch post named OpenRouter, Vercel and Cloudflare as gateway partners. That is a dated launch statement, not a present-day gateway inventory. Confirm the provider’s current model listing and exact model name before routing production traffic.
6. Use a Managed Enterprise Platform
xAI announced Grok 4.6 on Gemini Enterprise Agent Platform on August 21 and Microsoft Foundry on August 26. These are dated vendor announcements; confirm current tenant and catalog availability. Managed platforms may suit organizations that need governed endpoints or one catalog for evaluating several model vendors.
The August 12 xAI launch announcement did not name standard Grok.com consumer chat as a Grok 4.6 route. The retained evidence does not establish whether a selectable consumer chat mode has changed since; check xAI’s current product documentation for availability.
Grok 4.6 vs Grok 4.5
| Area | Grok 4.5 | Grok 4.6 |
|---|---|---|
| Launch date | July 8, 2026 | August 12, 2026 |
| Context window | 500K | 500K |
| Standard API price | $2 input / $6 output | $2 input / $6 output |
| AA Intelligence Index | 56 | 61 |
| DeepSWE v1.1 | 54.0% | 65.9% |
| Terminal-Bench v3.0 | 15.7% | 26.0% |
| Reasoning | Low, medium, high | Low, medium, high, xhigh |
| Primary launch emphasis | Coding, agentic tasks, knowledge work | Long-running agents, interactive and visual projects |
The earlier article comparison records the same 500K context and $2/$6 standard API rates for Grok 4.5 and Grok 4.6. This October 12 source check confirms the Grok 4.6 API context entry and the October 11 pricing capture; it does not recheck Grok 4.5’s values. xAI’s launch report emphasizes stronger agent and knowledge-work results for Grok 4.6 and adds an xhigh reasoning setting.
Kingy Verdict
Grok 4.6 is a credible frontier agent release. Matching GPT-5.6 Sol Max on xAI’s version of the Artificial Analysis Intelligence Index is attention-grabbing, but the more useful signal is breadth: Grok improves over 4.5 on every row xAI reports, with particularly large jumps on DeepSWE, Terminal-Bench, and APEX-Agents.
The launch does not prove Grok 4.6 is the best model for every coding or knowledge-work task. GPT-5.6 Sol Max and Fable 5 Max still post stronger results on several difficult agent evaluations, and xAI’s table mixes public and self-reported competitor numbers. Production teams should test completion rate, latency, token use, retries, and review time on their own work.
The price is attractive only if you quote it honestly. Below 200K prompt tokens, $2/$6 is compelling. At the long-context threshold, it becomes $4/$12 for the entire request. Grok 4.6’s real opportunity is therefore not “stuff 500K tokens into every prompt.” It is using a stronger long-horizon model with disciplined context, caching, and compaction.
FAQ
When was Grok 4.6 released?
Grok 4.6 was released on August 12, 2026.
How much does the Grok 4.6 API cost?
Below 200K prompt tokens, Grok 4.6 costs $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens. In xAI’s long-context band at 200K tokens, those rates double to $4, $1, and $12 respectively, and the higher rates apply to all tokens in the request.
What is the Grok 4.6 context window?
The xAI API model page lists a 500,000-token context window for Grok 4.6; Cursor lists 256K for its separate integration. The Grok 4.5 context figure in the comparison table is a historical article value and was not confirmed by this October 12 check.
What is the Grok 4.6 API model name?
The xAI API model page identifies the model string as grok-4.6. Check the current xAI API docs for endpoint support.
Is Grok 4.6 available in Cursor?
Yes. Cursor currently lists Grok 4.6 in its Models pool for individual and team plans. The first-week double-included-usage promotion has ended, so check the current allowance for your plan.
Does Grok 4.6 beat GPT-5.6 Sol?
There is no universal winner. In xAI’s launch table, Grok 4.6 matches GPT-5.6 Sol Max at 61 on the AA Intelligence Index and leads some knowledge-work evaluations, while GPT-5.6 Sol Max leads DeepSWE and Terminal-Bench by clear margins.
Related Kingy Reading
- Grok 4.5 Benchmarks: Pricing, Context, and the Opus Claim
- Best AI Coding Agent in 2026
- Cursor AI Tool Profile
- xAI Drops Grok Build