Breaking

SpaceXAI Releases Grok 4.6 for Long-Running Agents at $2 Input and $6 Output per Million Tokens

August 12, 2026 at 12:25 EDT

Foundation Models
AI Agents
Software Dev & Coding

SpaceXAI released Grok 4.6 on August 12, 2026, positioning the model as an upgrade for long-running agents, coding and knowledge work. It is available immediately through the xAI API, Grok Build, Cursor and several model gateways, with base pricing of $2 per million input tokens and $6 per million output tokens.

August 12, 2026 · SpaceXAI

Grok 4.6 is built for the long run

The new flagship focuses on sustained agent work, coding and ambitious interactive projects, with a 500,000-token context window and base API rates of $2 in and $6 out per million tokens.

Measured gain

DeepSWE v1.1 improves over Grok 4.5

Resolution rate · higher is better

An 11.9-point increase in SpaceXAI's launch evaluation, though GPT-5.6 Sol scored higher on the same test.

500K

token context window

per 1M input tokens

per 1M output tokens

2×

included usage in week one

Training path

From supplemental training to sustained work

Curated data

Reasoning, engineering and technical concepts

→

SFT trajectories

Regenerated with Grok 4.5 and filtered

→

Agentic RL

Coding, web, CAD and knowledge tasks

→

Longer trajectories

More self-testing and project refinement

What stands out

A broader first pass

Long-running agent and coding tasks
Interactive and visual application work
Self-testing on longer trajectories

What needs testing

Production reality

Launch benchmarks need independent replication
Cache misses can raise real input costs
Early access, moderation and reliability concerns remain

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.