The models
Use Grok 4.6 for code, chat, and agents. Use Imagine for pictures and video. Use Voice for speech. That is the official split as of August 2026.
Sources: the Grok 4.6 announcement, the 4.6 model page, and the models list.
Grok 4.6
Released 12 August 2026. Built on Grok 4.5 with a longer supplemental training run, regenerated SFT trajectories, and agentic RL across knowledge work, general coding, kernel work, web development, and CAD-like environments. SpaceXAI's own write-up says it stays with a task across many steps and produces a stronger first pass on visual and interactive projects than 4.5.
| Property | Value |
|---|---|
| API name | grok-4.6 |
| Context | 500,000 tokens |
| Knowledge cutoff | 1 February 2026 |
| Input | Text and images |
| Output | Text, no stated cap |
| Reasoning | low, medium, high (default), xhigh |
| Input price | $2 / 1M tokens (prompts under 200k) |
| Cached input | $0.50 / 1M |
| Output price | $6 / 1M tokens |
| Long-prompt rate | 2× the above when the prompt is ≥ 200k |
| Fast variant | 2× standard price |
APIs: Responses and Chat Completions. Tools: function calling, web search, X search, code execution. SpaceXAI recommends setting a prompt_cache_key (or x-grok-conv-id on Chat Completions) so multi-turn work actually hits the cache.
How it scores
From the 12 August 2026 announcement, Grok 4.6 High versus the previous flagship and two public peers. Best score in each row is not re-bolded here; these are the published figures.
| Eval | Grok 4.6 High | Grok 4.5 High | GPT-5.6 Sol Max | Fable 5 Max |
|---|---|---|---|---|
| AA Intelligence Index | 61 | 56 | 61 | 62 |
| GDPVal-AA v2 | 1753 | 1526 | 1728 | 1741 |
| CursorBench v3.2 | 69.9% | 66.7% | 67.2% | 70.5% |
| DeepSWE v1.1 | 65.9% | 54% | 73% | 70% |
| FrontierCode v1.1 (Extended) | 61.3% | 56.6% | 60.6% | 63.6% |
| APEX-Agents | 57.5% | 47.1% | 56.7% | 59.2% |
| Terminal-Bench v3.0 | 26% | 15.7% | 34.6% | 34.1% |
| APEX-SWE | 56.4% | 53.6% | — | 58.8% |
| AA-Briefcase | 1577 | 1313 | 1502 | 1574 |
| Harvey LAB (Vals) | 15.8% | 12.9% | 2.5% | 11.3% |
SpaceXAI says 4.6 matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index, a composite of nine benchmarks.
The rest of the text line
Still on the public price sheet, for work that does not need the new flagship:
| Model | Context | Input / 1M | Output / 1M |
|---|---|---|---|
grok-4.5 |
500k | $2.00 | $6.00 |
grok-4.3 |
1M | $1.25 | $2.50 |
grok-4.20-0309-reasoning |
1M | $1.25 | $2.50 |
grok-build-0.1 |
256k | $1.00 | $2.00 |
Long-context rows (prompt ≥ 200k) are 2× those input and output rates. Cached input is cheaper; see the models page.
Media and voice
| Model | Meter |
|---|---|
grok-imagine-image-2.0 |
$0.04 / image |
grok-imagine-image |
$0.02 / image |
grok-imagine-image-quality |
$0.05 / image |
grok-imagine-video-1.5 |
$0.080 / sec |
grok-imagine-video |
$0.050 / sec |
grok-voice-think-fast-2.0 |
$0.08 / min audio + $0.004 / text input |
Aliases
<name> tracks the latest stable. <name>-latest tracks the latest including newer cuts. <name>-<date> is a pinned snapshot. For most integrations, the unadorned name is the one that should be in the config.
from xai_sdk import Client
from xai_sdk.chat import user
import os
client = Client(api_key=os.getenv("XAI_API_KEY"))
chat = client.chat.create(model="grok-4.6")
chat.append(user("Explain quantum computing"))
print(chat.sample().content)