ChinaModelAPI

News / Model Watch · Western frontier

Western watch Grok 4.6 xAI Frontier baseline
2026-08-13 · Model Watch

Grok 4.6: xAI's New Recommended Flagship, Now in the API

xAI's official developer documentation now lists grok-4.6 as its latest model and recommends it for code. We track Grok as a Western frontier baseline against our Chinese open models (the same role GPT-5.6 Sol and Claude Opus 5 play) — so here is a clean, sourced read on what the docs actually say, and what it means for a China-model stack.

Direct answer

Grok 4.6 is now xAI's recommended flagship in the official API docs. On docs.x.ai/docs/models, grok-4.6 is listed with a 500k context window at $2 / $6 per 1M tokens (under 200k prompt) — and the docs tell developers “for everything else, including code, use Grok 4.6,” calling it xAI's most intelligent and fastest model. Knowledge cutoff is February 1, 2026. It is also selectable inside xAI's coding tooling (the grok-build model family). Note: Grok is a Western closed model, not routed by ChinaModelAPI — we cover it as a comparison baseline only.

What xAI's official docs say (verified)

We use official documentation as the source of truth, not social chatter. As of 2026-08-13, the xAI developer models page lists the following for Grok 4.6:

SpecGrok 4.6 (per xAI docs)
Model IDgrok-4.6
Context window500k tokens
Price (< 200k prompt)$2.00 / 1M input · $6.00 / 1M output
Price (≥ 200k prompt)$4.00 / 1M input · $12.00 / 1M output
Knowledge cutoffFebruary 1, 2026
Positioning“Most intelligent and fastest” · recommended for code

Parameter count and training detail come from Elon Musk's X announcements (1.5T parameters, improved SFT & RL) — treated as secondary discovery, not the primary docs source.

From the Aug 7 target to a docs listing

Grok 4.6 had a noisy run-up. Elon Musk announced on X (late July) that Grok 4.6 would ship “around August 7” as a 1.5T model with significantly improved supervised fine-tuning and reinforcement learning, succeeding the “Opus-class” Grok 4.5. That date slipped, mainstream coverage stayed conflicting, and the model briefly surfaced in third-party coding tools before settling. The clean signal that matters to builders: grok-4.6 is now present in xAI's own API documentation with pricing and a usage recommendation.

  • Lineage: successor to Grok 4.5 (released July 2026, positioned as an “Opus-class” model).
  • Coding tooling: the grok-build-0.1 model in the same docs page underpins xAI's coding CLI — consistent with developers seeing Grok 4.6 inside xAI's building tools.
  • Next: Grok 4.7 (a larger ~2.1T model) has been teased for the weeks ahead — not yet in scope here.

How this fits a China-model builder

Grok 4.6 is not a model ChinaModelAPI routes — it is a Western closed flagship. We track it as a quality and cost ceiling reference, exactly like GPT-5.6 Sol and Claude Opus 5 in our model-watch tables. The practical use is comparison, not integration.

ModelOrigin / accessBest role in your stack
Grok 4.6xAI · closed APIWestern quality/cost ceiling reference
DeepSeek-V4-Pro-0813DeepSeek · open weightsFrontier agents/reasoning, self-hostable, cheap
Qwen3.8-MaxAlibaba · open weightsCoding & long-horizon work
GLM-5.2 / Kimi K3Zhipu / Moonshot · open1M-context long-horizon & agents

Full live comparison: Chinese AI model comparison.

Primary sources

FAQ

Can I call Grok 4.6 via ChinaModelAPI?

No. Grok is a Western closed model. ChinaModelAPI routes Chinese frontier models (DeepSeek, Qwen, GLM, Kimi, Seedance, HappyHorse). We list Grok 4.6 here only as a Western baseline for comparison.

Is there a launch blog post?

We did not separately confirm a standalone Grok 4.6 launch post; the authoritative signal we rely on is the official API docs listing grok-4.6 with pricing and a usage recommendation.

500k vs 1M context?

Grok 4.6 offers 500k per the docs — large, but below the 1M context several Chinese flagships (DeepSeek-V4-Pro, Qwen3.8, GLM-5.2, Kimi K3) advertise.

Cheaper Western option?

For cost-sensitive volume, open-weight Chinese models (DeepSeek-V4-Flash, Qwen/GLM/Kimi tiers) are typically far cheaper per token than Grok's $2/$6 band.