Status on September 4, 2026: Grok 4.7 is not officially announced
SpaceXAI’s current news page has no Grok 4.7 launch post, while the model documentation still marks Grok 4.6 as the new flagship. API release notes updated on September 2, 2026 also do not introduce a 4.7 model. [1][3][4]
The latest confirmed model is Grok 4.6
SpaceXAI released Grok 4.6 on August 12, 2026. It is positioned for coding, agentic tasks, and knowledge work, with an emphasis on long-running processes and ambitious interactive and visual projects. [2][4]
Official Grok 4.6 specifications
| Property | Grok 4.6 | Grok 4.7 |
|---|---|---|
| Status | officially available | unconfirmed |
| Context | 500K tokens | no data |
| Input | text + image | no data |
| Output | text | no data |
| Reasoning levels | low / medium / high / xhigh | no data |
The documentation lists a 500,000-token context window, text and image input, text output, and no stated text output limit. Reasoning effort supports low, medium, high — the default — and xhigh. [3]
Pricing: 4.6 is documented, 4.7 is not
| Prompt range | Input / 1M | Cached / 1M | Output / 1M |
|---|---|---|---|
| < 200K tokens | $2.00 | $0.50 | $6.00 |
| ≥ 200K tokens | $4.00 | $1.00 | $12.00 |
For Grok 4.6, official pricing distinguishes shorter and longer context. Below 200K prompt tokens, rates are $2 per 1M input tokens, $0.50 per 1M cached tokens, and $6 per 1M output tokens. At 200K tokens and above, rates rise to $4/$1/$12. No Grok 4.7 pricing has been published. [4][5]
Grok 4.6 tools and APIs
Grok 4.6 supports the Responses API and Chat Completions, plus function calling, web search, X search, and code execution. SpaceXAI recommends a conversation cache key to make cache hits more reliable in long sessions. [3]
Grok 4.6 benchmarks: vendor-reported results
| Benchmark | Grok 4.6 High | Grok 4.5 High |
|---|---|---|
| AA Intelligence Index | 61 | 56 |
| GDPVal-AA v2 | 1753 | 1526 |
| CursorBench 3.2 | 69.9% | 66.7% |
| DeepSWE 1.1 | 65.9% | 54.0% |
| FrontierCode 1.1 Extended | 61.3% | 56.6% |
In its launch material, SpaceXAI reports 61 on the AA Intelligence Index, 1753 on GDPVal-AA v2, 69.9% on CursorBench 3.2, and 65.9% on DeepSWE 1.1 for Grok 4.6 High. These are vendor-reported results for a specific setup, not data for a future Grok 4.7. [2]
Independent reference point: Artificial Analysis
Artificial Analysis also scored Grok 4.6 High at 61 on its Intelligence Index and described the model as returning SpaceXAI to the intelligence frontier at relatively low token pricing. That validates the position of 4.6, not any claim about 4.7. [11]
Grok 4.6 partner availability
After launch, Grok 4.6 reached GitHub Copilot, Amazon Bedrock, Gemini Enterprise Agent Platform, and Microsoft Foundry. Those rollouts happened between August 14 and August 26, 2026. [6][7][8][9]
What cannot be attributed to Grok 4.7 today
There is no confirmed `grok-4.7` model ID, release date, context window, price, output limit, reasoning levels, supported tool list, or benchmark table. Any concrete number in those areas would currently be speculation. [1][3][4][5]
Fast version cadence is not evidence of a 4.7 launch
Grok 4.5 was announced on July 16, 2026 and Grok 4.6 on August 12, 2026. That short gap does not establish a schedule for the next release; SpaceXAI has not published a Grok 4.7 roadmap date. [2][10]
How to verify a real Grok 4.7 release
A trustworthy confirmation should include a SpaceXAI news post, an entry in the model catalog, an API release-note update, and a documented working API slug. A “Grok 4.7” label in a third-party interface should not replace first-party documentation. [1][3][4]
What to verify immediately after release
Once an official release exists, the key items are model ID, context window, short- and long-context pricing, caching, reasoning levels, tools, modalities, API regions, rate limits, benchmark methodology, and any breaking changes from 4.6. [3][5]
- Model ID and API aliases
- Context and output limit
- Input/cache/output pricing
- Reasoning effort
- Web/X search, code execution, and function calling
- Vendor and independent benchmarks
- Regional and partner availability
- Migration changes from Grok 4.6
Should teams prepare a 4.6 → 4.7 migration now?
There is nothing to migrate to yet. Production systems should use documented model identifiers and evaluate a new version only after its API behavior, pricing, and documentation are published. [3][4]
What this means for users and companies
If the goal is to use the latest confirmed Grok for coding, agents, and knowledge work, the current documented choice is Grok 4.6. Teams can build their own evaluation suite now and rerun it when an official successor appears. [2][3][11]
Bottom line: do not publish a specification that does not exist
As of September 4, 2026, a factual Grok 4.7 article must begin by saying the model has not been officially introduced. Grok 4.6 is a useful baseline, but its specifications should not be copied forward or used to predict benchmark gains without sources. [1][3][4]
