Grok 4.6 Release Watch: Date, Status & API Availability
Grok 4.6 is not publicly released as of July 30, 2026. Elon Musk says it is expected in about one week, pointing to roughly August 6-7 if the estimate holds. The model is not yet listed in xAI's public model catalog or API release notes, so there is no verified public model ID, price, context window, or EvoLink route.
The important distinction is simple: a public release estimate is not the same as a callable API release. Developers should treat Grok 4.6 as an official availability watch, not as a production dependency.
Grok 4.6 Status at a Glance
| Question | Status on July 30, 2026 | What it means |
|---|---|---|
| Has Grok 4.6 been publicly named? | Yes | Elon Musk has referred to the model directly |
| Is there an expected release window? | Yes, about one week | Roughly August 6-7 if the estimate holds |
| Is Grok 4.6 listed in xAI's model catalog? | No | No public API model record is available |
| Is it in xAI's API release notes? | No | Provider API availability is not documented |
| Is there a verified model ID? | No | Do not assume grok-4.6 is callable |
| Is official pricing published? | No | Grok 4.5 pricing is not a Grok 4.6 quote |
| Is Grok 4.6 available through EvoLink? | Not verified | No EvoLink route should be claimed before testing |
Is Grok 4.6 Released?
No. The current evidence supports an expected near-term release, not a completed public launch.
For developers, release should be checked across four separate states:
- Publicly announced: the provider names the model or gives a release estimate.
- Consumer access: the model appears in Grok, X, or another end-user product.
- Provider API access: xAI publishes a callable model record and commercial terms.
- EvoLink access: EvoLink verifies the route, model identity, pricing, and request behavior.
These states may happen together, but they should not be assumed to.
“Released” Can Mean Six Different Things
A single “Grok 4.6 is out” headline will not answer whether a developer can use it in production. Each surface needs its own evidence:
| Release surface | Evidence that changes the status | What it does not prove |
|---|---|---|
| xAI announcement | A dated xAI post, model card, or release note | That any account can call the model |
| Grok consumer product | Grok 4.6 appears in the model picker or an account rollout | API access, pricing, or stable capacity |
| Grok Build or partner IDE | A named model is selectable in a developer tool | A generally available xAI API model ID |
| xAI API | Official model record, commercial terms, and a successful eligible request | Availability in every region or service tier |
| Regional availability | Successful access from the target account and region | Access for another account, contract, or geography |
| EvoLink gateway | Verified upstream identity plus a tested EvoLink route | That every existing Grok 4.5 request is migration-compatible |
This distinction is especially important because xAI has previously separated product, API, and regional milestones. Its release notes list Grok 4.5 API availability on July 8 and a separate EU availability update on July 17. The Grok 4.6 tracker should therefore record channel and region explicitly, not collapse every milestone into one release date.
Expected Grok 4.6 Release Date
Best current estimate: August 6-7, 2026. This is a release estimate, not a confirmed general-availability date.
The distinction matters because an announcement can refer to several different events:
- access in the Grok consumer app;
- availability in Grok Build;
- a limited account rollout;
- xAI API availability;
- a broader regional release.
Until xAI publishes the access channel, model record, and release note, developers cannot safely infer which event the estimate describes.
How the Grok 4.6 Timeline Changed
Public reporting has moved quickly:
- July 18: early reporting described a larger successor to Grok 4.5 and focused on training progress.
- July 24: Musk said Grok 4.6 was about two weeks away.
- July 30: Musk shortened the estimate to about one week.
This progression increases confidence that a release is near. It does not establish the final API date, regional scope, or commercial terms.
The same caution applies to model size. Early coverage described Grok 4.6 as a 2-trillion-parameter model. Newer coverage has described a 1.5-trillion-parameter Grok 4.6 with stronger supervised fine-tuning and reinforcement learning, while assigning a larger scale to Grok 4.7. xAI has not published a Grok 4.6 model card that resolves that conflict.
Confirmed and Unconfirmed Grok 4.6 Details
The most useful release tracker separates durable facts from claims that still need provider documentation.
Confirmed or directly stated
- The name Grok 4.6 has been used publicly by Elon Musk.
- Musk gave an expected release estimate of about one week on July 30, 2026.
- Grok 4.5 is the documented predecessor and current xAI API flagship.
Not yet confirmed by xAI documentation
| Field | Current evidence state |
|---|---|
| Exact release date and time | Not published |
| Consumer rollout channels | Not published |
| xAI API availability | Not listed |
| API model ID | Not listed |
| Input and output pricing | Not listed |
| Context window | Not listed |
| Modalities | Not listed |
| Reasoning controls | Not listed |
| Regions and account eligibility | Not listed |
| Benchmarks | Not published |
| Parameter count | Conflicting secondary reports |
| EvoLink route | Not verified |
This table is intentionally incomplete. Filling unknown cells with Grok 4.5 values would make the page look more complete while making it less reliable.
Grok 4.6 API Availability, Model ID, and Pricing
There is no documented Grok 4.6 API record as of July 30.grok-4.5, a 500,000-token context window, and prices of $2 per million input tokens and $6 per million output tokens. Those are verified Grok 4.5 facts. They must not be reused as Grok 4.6 specifications.Before treating Grok 4.6 as API-live, verify all of the following:
| Verification | Evidence developers should require |
|---|---|
| Model identity | An official model record and the model returned by the API |
| Access | A successful request from an eligible account and region |
| Pricing | Provider terms tied to the exact model and service tier |
| Request behavior | Supported endpoint, reasoning controls, tools, streaming, and structured output |
| Limits | Context, output, rate limits, and regional restrictions |
| Production behavior | Error rate, latency, retry behavior, and capacity under representative traffic |
EvoLink should only describe a Grok 4.6 route as available after the same identity, pricing, and request checks pass through its gateway.
The Grok 4.5 compatibility baseline that 4.6 must re-pass
Model access is not only a question of whether one prompt returns a response. The current xAI documentation exposes several Grok 4.5 behaviors that can affect an existing integration:
| Current Grok 4.5 behavior | Why it matters | Grok 4.6 release-day check |
|---|---|---|
reasoning_effort supports low, medium, and high; high is the documented default | A default change can alter latency and output-token use without an application-code change | Replay each supported effort level and compare usage plus latency |
| Reasoning cannot be disabled | A “fast” test still needs to account for reasoning behavior | Confirm the minimum supported effort and response shape |
presencePenalty, frequencyPenalty, and stop are not supported with reasoning models | Existing generic OpenAI-compatible request builders may send fields that cause errors | Test the exact production payload, not a simplified sample |
Usage reports reasoning_tokens | Cost and efficiency analysis needs more than visible output length | Verify that usage fields remain available and correctly logged |
| Built-in tools and function calling have separate execution behavior | A successful text response does not prove search or agent compatibility | Test every tool class used by the workload |
| Structured outputs support JSON Schema, with some keywords documented as best effort | Schema acceptance can fail even when prose quality improves | Run the production schemas and validate them independently |
| Repeated prompt prefixes may use automatic prompt caching | Cache misses can change time to first token and effective cost | Compare cold and warm requests using the production prefix pattern |
These are Grok 4.5 documentation facts, not Grok 4.6 promises. They define the minimum compatibility surface to retest when 4.6 appears.
What Could Stronger Post-Training Change?
Recent reporting frames Grok 4.6 as a post-training improvement rather than a simple parameter-count increase. If that description is accurate, the upgrade should be judged through behavior, not model size.
The most relevant outcomes would be:
- fewer instruction-following failures on long tasks;
- more reliable tool selection and argument construction;
- fewer unproductive reasoning loops;
- better recovery after a tool or execution error;
- higher first-pass acceptance on coding and knowledge-work tasks;
- lower token use per successful task.
These are test hypotheses, not confirmed Grok 4.6 capabilities. xAI has not published matched Grok 4.5 and Grok 4.6 results.
The First 24-Hour Grok 4.6 Verification Runbook
If Grok 4.6 appears, a useful first-day process separates documentation checks from production routing:
| Time after a credible release signal | Verification work | Exit condition |
|---|---|---|
| 0-30 minutes | Capture the official announcement, model page, model ID, rollout channel, date, and region | The release surface is identified without inference |
| 30-90 minutes | Verify price, context, limits, supported endpoints, reasoning controls, streaming, tools, and structured output | Commercial and request contracts are documented |
| 1.5-4 hours | Replay a fixed set of successful, failed, tool-using, and schema-constrained Grok 4.5 traces | Blocking compatibility failures are understood |
| 4-12 hours | Run a reversible internal canary with returned-model, latency, usage, error, retry, and fallback logs | The route behaves predictably under limited traffic |
| 12-24 hours | Compare accepted-task quality and total execution cost; record region and service-tier caveats | A workload-specific test/hold/expand decision is made |
One successful curl request is enough to prove reachability, not production readiness. The first public update can report the model as API-callable while still marking pricing, compatibility, capacity, regional access, or EvoLink support as pending.
What Developers Can Do Before the Release
Teams can prepare without hard-coding an unverified model ID:
- Keep model selection in configuration rather than application logic.
- Save 20-50 representative Grok 4.5 traces, including successful tasks and expensive failures.
- Define acceptance checks for correctness, tools, latency, retries, and cost per accepted task.
- Separate offline replay, canary, and default-route stages.
- Keep a tested fallback until Grok 4.6 passes the workload's hard gates.
- Watch the xAI model catalog and release notes instead of relying on copied launch posts.
What Will Trigger an Update to This Page?
This release watch should be updated when one of these records changes:
- xAI publishes a Grok 4.6 announcement or model card;
- the model appears in xAI's model catalog;
- xAI publishes a callable model ID;
- pricing and regional access become visible;
- EvoLink verifies a route and successful request behavior;
- reproducible Grok 4.6 evaluations become available.
FAQ
When will Grok 4.6 be released?
Elon Musk said on July 30, 2026 that Grok 4.6 would come out in about one week. That points to roughly August 6-7 if the estimate holds, but xAI has not published a fixed general-availability date.
Is Grok 4.6 released now?
No. It is not listed in xAI's public model catalog or API release notes as of July 30, 2026.
Is Grok 4.6 available through the xAI API?
Not according to the current public documentation. Developers should wait for an official model record, model ID, pricing, and a successful eligible request.
Is Grok 4.6 available on EvoLink?
No verified EvoLink route has been announced. EvoLink can prepare for support, but availability should only be claimed after the upstream route and EvoLink request behavior are confirmed.
What is the Grok 4.6 API model ID?
grok-4.6 is a valid request value until it appears in official documentation.Does Grok 4.6 have 1.5T or 2T parameters?
Public reporting conflicts. Early sources described 2T, while newer coverage described 1.5T and assigned a larger model to Grok 4.7. xAI has not published a model card that resolves the difference.
How much will Grok 4.6 cost?
Pricing is unknown. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens through xAI, but those rates should not be presented as Grok 4.6 pricing.
Should developers wait for Grok 4.6?
Sources
- Elon Musk: Grok 4.6 release estimate
- xAI model catalog
- xAI API release notes
- xAI: Introducing Grok 4.5
- xAI reasoning models documentation
- xAI tools overview
- xAI structured outputs documentation
- xAI prompt caching documentation
- NewsBytes: Grok 4.6 and Grok 4.7 reporting
- IT之家: July 24 Grok 4.6 timeline report
