Grok 4.6 is xAI and Cursor's August 12, 2026 model for long-running agents and interactive visual work. Published API price is $2 per million input tokens and $6 per million output tokens. A Fast variant costs twice that. o10 lists Grok 4.6 at band C3 on xAI and OpenRouter.
Last updated: 2026-08-14. Full catalog. Reference prices are dated snapshots. Compare total cost and measured task quality before choosing a route.
Grok 4.6 is an xAI frontier model released 12 August 2026 and trained jointly with Cursor. It targets long-running agents, multi-step coding, and stronger first passes on interactive and visual projects than Grok 4.5. It is live in Cursor, Grok Build, the xAI API, and partners including OpenRouter, Vercel, and Cloudflare.
xAI launch, 12 August 2026.
Cursor markets Grok 4.6 as a joint Cursor and SpaceXAI (xAI) model. Cursor says the pair ran a longer supplemental training pass on Grok 4.5 with engineering data and an improved optimizer. In Cursor it sits in the Cursor Models pool with Grok 4.5 and Composer 2.5. The API identity remains grok-4.6 from xAI.
Cursor Grok 4.6 model card.
Grok 4.6 list price is $2 per million input tokens, $0.50 per million cached input, and $6 per million output tokens. The Fast variant is $4 / $1 / $12. xAI offered 2x included usage in Cursor and Grok Build for the first week after launch. Endpoint prices on OpenRouter or Vercel can differ from list.
xAI and Cursor list prices, August 2026.
xAI reports Grok 4.6 at 61 on the Artificial Analysis Intelligence Index, tied with GPT-5.6 Sol and just behind Fable 5 Max at 62. CursorBench v3.2 is 69.9%, DeepSWE v1.1 is 65.9%, FrontierCode v1.1 Extended is 61.3%, and Terminal-Bench v3.0 is 26%. Those figures are vendor-published; run your own evals before enforce.
xAI Grok 4.6 eval table, 12 August 2026.
Grok 4.6 runs in Cursor (desktop, web, iOS, CLI, SDK), Grok Build, the xAI API as grok-4.6, and aggregators such as OpenRouter (x-ai/grok-4.6), Vercel, and Cloudflare. o10 fulfillment maps grok-4-6 to xAI and OpenRouter upstream IDs.
o10 fulfillment overlay 2026-08-14.
Grok 4.6 is band C3: frontier-capable, not the cheapest chat default. o10 will only route it when the use-case eval floor needs that class. Shadow mode can prove whether Grok 4.6, Gemini 3.7 Flash, or a cheaper C1/C2 model clears the same bar.
o10 band C3.
Grok 4.6 in the o10 unified catalog.
| Field | Value |
|---|---|
| o10 slug | `grok-4-6` |
| Type | text |
| Band | C3 |
| Routable | Yes |
| Source provider | xai |
| Primary host | xAI |
| Gateway ID | `x-ai/grok-4.6` |
| Context window | 500,000 tokens |
| Released | 2026-08-12 |
Gateway snapshot pricing for Grok 4.6.
| Unit | Price |
|---|---|
| Input | $2/1M tokens |
| Output | $6/1M tokens |
Upstream venues configured for Grok 4.6.
| Host | Upstream model ID |
|---|---|
| xai | xai/grok-4.6 |
| openrouter | openrouter/x-ai/grok-4.6 |
Eval-gated model selection for this slug.
o10 slug: grok-4-6. Band C3 (frontier-capable). Routable. Prefer o10/frontier when the quality floor needs a flagship agent model; pin grok-4-6 when Cursor/xAI is the required venue.
See [/models/routing](/models/routing) for smart aliases (o10/auto, o10/frontier, o10/squad) and [/kyi](/kyi) for governance.
Post-training refresh aimed at agents that stay on a task for many steps.
xAI reused the Grok 4.5 base and ran a longer supplemental training pass, then regenerated SFT trajectories with Grok 4.5 across reasoning efforts, agent harnesses, STEM, software engineering, and knowledge work. RL covered kernel optimization, web development, and CAD-style environments.
The product claim is not a bigger chat model. It is fewer abandoned trajectories: research, structure an app, implement interactions, self-test, then iterate. That is why Cursor co-shipped it for agentic coding.
List prices and launch claims from xAI (12 August 2026) and Cursor's Grok 4.6 model card. Context window 500K is reported by public catalogs; confirm against xAI docs. o10 overlay snapshot 2026-08-14.
Grok 4.6 is $2/1M input and $6/1M output at xAI list price. Fast is twice that. Cached input is $0.50/1M on the standard variant. Verify the venue you actually call.
Public catalogs report a 500,000-token context window for Grok 4.6. xAI's launch note does not state the figure; confirm against current xAI API docs before designing long-context jobs.
Yes. Cursor shipped Grok 4.6 on launch day across desktop, web, iOS, CLI, and SDK, with 2x included usage for the first week. It is not a Cursor-only weight: the same model is on the xAI API.
On xAI's table, Grok 4.6 High lifts AA Intelligence from 56 to 61, CursorBench from 66.7% to 69.9%, and Terminal-Bench v3.0 from 15.7% to 26%. Gains are post-training on the Grok 4.5 base, not a new pretrain.
Send model "grok-4-6" to https://app.o10.io/v1, or use o10/frontier and let evals decide. Upstream IDs: xai/grok-4.6 and openrouter/x-ai/grok-4.6.
o10 routes Grok 4.6 only when per-use-case evals clear your quality floor at band C3. If a cheaper model passes, auto will not spend C3. See /models/routing.
o10 slug "grok-4-6"; xAI/gateway id "x-ai/grok-4.6"; OpenRouter "x-ai/grok-4.6". Snapshot 2026-08-14.