Model
Model explorer

Grok 4.6

CLOSED
SpaceXAI · Grok 4.6 family · released Aug 12, 2026

SpaceXAI's August 12, 2026 successor to Grok 4.5, 35 days later, focused on long-running agents and interactive/visual work. Same headline $2/$6 per Mtok as 4.5, but cached input rose from $0.30 to $0.50/Mtok, so a cache-heavy prompt costs MORE against the newer model; a 'fast' variant is double price. Available in Cursor, Grok Build, the SpaceXAI API, OpenRouter, Vercel and Cloudflare. SpaceXAI's claim is parity rather than a lead: it matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61 each, with Claude Fable 5 at 62). All figures recorded here are the Grok 4.6 High column of SpaceXAI's own eval table, with competitor rows drawn by SpaceXAI from those developers' published results. The table is honest about losses — Terminal-Bench 3.0 26.0 against GPT-5.6 Sol Max 34.6 and Fable 5 Max 34.1, DeepSWE 65.9 against 73.0 and 70.0 — while its clearest wins are GDPval-AA v2 (1753) and Harvey LAB (15.8, far ahead of Sol's 2.5).

ReasoningCodingVisionFunction callingTool useAgentic
2960.9
Elo · rank #9
Parameters
Undisclosed
Active params
Undisclosed
Context
500K tokens
Architecture
Mixture-of-Experts (size undisclosed); a longer supplemental training run over the Grok 4.5 foundation with regenerated SFT trajectories and agentic RL across kernel optimization, web development and CAD environments
License
Proprietary
Languages
API price (in/out)
$2 / $6
Modalities
text · vision
Benchmark results
Bar shows position within the tracked field; marker = field best
AA-BriefcaseAgents1577#3
best: Claude Opus 5 · 1720
CursorBenchCoding69.9%#4
best: Claude Fable 5.1 · 73.4%
DeepSWECoding65.9%#8
best: Muse Spark 1.3 · 75.4%
best: Claude Fable 5 · 64.9%
GDPval-AAAgents1753#7
best: Claude Fable 5 · 1932
best: this model · 15.8%
Terminal-Bench 3.0Agents26.0%#4
best: GPT-5.6 Sol · 34.6%
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
$2
Output / M tok
$6
Grok 4.6 family
Elo progression across releases
API price $2/$6 · each benchmark row carries its own source badge (see methodology)