xAI Grok
Listed at https://console.x.ai
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Grok 4.6: $2/M tokens (fast variant $4/M) | per million tokens |
| Output | Grok 4.6: $6/M tokens (fast variant $12/M) | per million tokens |
🤖 Supported Models (4)
✨ Pros
- ✓Grok 4.6 (Aug 12) matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61)
- ✓Built for long-running agents: multi-step research, coding, turning ideas into working artifacts
- ✓Priced $2/$6 per M tokens, roughly half the cost of comparable frontier models
- ✓Fast variant at 2x price for low-latency workloads
- ✓OpenAI-compatible API; available in Cursor, Grok Build, and via OpenRouter, Vercel, Cloudflare
⚠️ Cons
- ×Near-frontier but slightly behind Fable 5 Max on several evals (AA Intelligence 61 vs 62)
- ×No multimodal input support in this release (positioned for knowledge work and agents)
- ×Proxy required for China access
- ×Platform ecosystem (SDKs, tooling) less mature than OpenAI/Anthropic
- ×No official free API credit at launch
🎯 Best For
Long-running agents and coding (codebase work, multi-step research); developers who want near-frontier quality on a tighter budget; real-time X/Twitter data integration
💰 Pricing & Plans
| Model | Input ($/M) | Output ($/M) | Notes |
|---|---|---|---|
| Grok 4.6 | $2.00 | $6.00 | Flagship, agentic + coding focus (Aug 12, 2026) |
| Grok 4.6 Fast | $4.00 | $12.00 | Fast variant at 2x price for low latency |
| Grok 4.5 | $2.00 | $6.00 | Prior flagship, still served |
| Grok 4.5 Fast | $4.00 | $12.00 | Fast variant of prior generation |
🔧 API & Developer Experience
- •API Surface: OpenAI-compatible HTTP API at console.x.ai (chat completions, streaming, function calling). The xAI SDK and OpenAI SDK both work, so migration from OpenAI is near-zero.
- •Models: Grok 4.6, Grok 4.6 Fast, plus prior Grok 4.5 / Grok 4.5 Fast. The flagship is tuned for agentic RL across knowledge work, general coding, kernel optimization, web dev and CAD.
- •Availability: Live today in Cursor and Grok Build (2x included usage for the first week), in the API, and through partners OpenRouter, Vercel and Cloudflare.
- •Agentic focus: Stronger first passes on visual/interactive projects and self-testing on longer trajectories; Grok 4.6 checks its own work before moving on, useful for building apps end-to-end.
- •Tools & SDK: x.ai CLI installer (curl https://x.ai/cli/install.sh), full REST docs at docs.x.ai, and integrations with Cursor / Grok Build.
- •No free API tier: Grok 4.6 has no official free API credit at launch; try it free inside Grok Build, then pay-as-you-go via the API.
🚀 Agent-Ready Reasoning & Coding
Grok 4.6's headline is long-running-agent capability: it can research unfamiliar domains, structure an application, implement the core interactions, and keep refining through several rounds of feedback. On official evals it scored GDPVal-AA v2 1753, CursorBench v3.2 69.9%, DeepSWE v1.1 65.9%, and FrontierCode v1.1 (Extended) 61.3%, matching GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61 vs 61) and trailing Fable 5 Max by a single point (62). xAI says Grok 4.6 does a longer supplemental training run with curated model-generated reasoning data and an improved RL recipe, which materially strengthens first-pass output on visual and interactive builds. For a frontier-adjacent model at roughly half the per-token price of rivals, this is the strongest agentic-coding value xAI has shipped to date.
🌐 Regional Availability & Latency
Grok 4.6 is served from xAI's multi-region inference fleet in the US and EU (API console.x.ai), with a dedicated Fast variant priced at 2x that targets low-latency interactive and agent-loop workloads. There is no mainland-China region: direct access from China requires a stable overseas proxy or an aggregator (OpenRouter, Vercel AI Gateway, Cloudflare) that fronts the xAI API. Latency-wise, the standard Grok 4.6 trades a little speed for frontier reasoning; teams needing the quickest first-token time should select the Fast model. Business tiers add dedicated capacity and higher rate limits through xAI sales. For China-based developers, the practical route is an OpenAI-compatible aggregator that can route around regional blocks.