Coding agents
Near frontierAA Coding Agent Index (Grok Build)
Composite of DeepSWE, Terminal-Bench v2, SWE-Atlas QnA. Grok 4.5 is 2nd, on par with GPT-5.5 in Codex.
SPIRITT
Grok 4.5Grok 4.5 is the cheap smart workhorse for agent work
SpaceXAI built Grok 4.5 to excel at coding, agentic tasks, and knowledge work: strong tool use, high token efficiency, and near-frontier intelligence at a lower cost. On SPIRITT, you run it inside a fully equipped cloud environment with tools, browser, and durable memory.
Launch week on X
Real posts from people testing Grok 4.5: speed, cost, coding agents, and browser use.
Official launch card
From SpaceXAI: Grok 4.5 is the smartest Grok model yet, built to excel at coding, agentic tasks, and knowledge work. Trained with heavy RL on multi-step software engineering and technical work, with Cursor data in supplemental training.
What it was built for
Agentic coding loops, terminal work, knowledge work, and long multi-step jobs where token efficiency matters. SpaceXAI positions it as near-frontier intelligence with much lower cost and output volume than premium peers.
Independent reads put Grok 4.5 fourth on the AA Intelligence Index and on par with GPT-5.5 in Codex on the Coding Agent Index inside the Grok Build harness, at a fraction of the cost per task.
Agentic codingKnowledge workToken efficiencyGrok Build500k contextVision input
AA Intelligence Index
Source: Artificial Analysis (Jul 8, 2026). Grok 4.5 at 54, +16 vs Grok 4.3. Charts below show where it is strong for agent work.
Where Grok is strong vs other models
Same score means nothing alone. These are head-to-head charts so you can see where it leads, where it is close, and what that means for real agent work.
Terminal agents
Terminal-Bench 2.1
Agentic terminal coding. Grok is effectively tied with the leaders.
Software engineering
SWE-Bench Pro
Diverse software engineering resolve rate. Competitive mid-pack, strong value.
Long-horizon coding
DeepSWE 1.0
Long-horizon agentic coding. Grok sits just under the top cluster.
Agent cost
WinnerCoding Agent cost per task
Lower is better. Grok Build is much cheaper than Claude Code / Codex for similar Coding Agent Index scores.
Intelligence cost
WinnerCost per AA Intelligence task
Lower is better. Grok is on the performance/cost Pareto frontier.
Knowledge reliability
WinnerAA-Omniscience Index
Accuracy rose, but hallucination rate also rose. Stronger knowledge with more confidence risk.
API pricing
WinnerOutput $ per 1M tokens
Headline API pricing comparison. Grok undercuts premium peers on output tokens.
Sources: Artificial Analysis public writeups (Jul 8, 2026) and SpaceXAI launch materials. Grok 4.5 is near-frontier on agentic coding economics, not the absolute top of every pure-coding board. On SPIRITT, you do not just call an API. You run Grok inside a workspace with tools, memory, and a computer already set up.
How It Works
From zero to Grok 4.5 running real work in a cloud agent environment
01
Open a workspace
Start free, including trial. You land in a fully equipped cloud computer: browser, files, terminal, integrations, and memory. No local setup. No thin chat box pretending to be an agent.
02
Pick Grok 4.5
Open the model picker and choose Grok 4.5. Same model shipped for agentic work, now inside a workspace that already has tools, browser control, and durable context. Available across plans, including trial.
03
Build or automate
Tell it what to ship or what to run. Grok can code, call tools, drive the browser, coordinate multi-step work, and keep going while you step away. The point is not another chat window. It is an agentic environment where Grok actually does the job.
Try Grok 4.5 in an agentic environment
A real cloud environment, model picker, tools, and memory. Included in all plans, including trial.
Questions
Frequently asked questions
01What is Grok 4.5?+
Grok 4.5 is SpaceXAI's frontier model built for coding, agentic tasks, and knowledge work. It emphasizes speed, token efficiency, and real multi-step agent loops, and you can run it inside SPIRITT Workspaces with tools, browser, and a full cloud computer.
02How is Grok 4.5 different from previous generations?+
Grok 4.5 is a large jump over Grok 4.3 on intelligence and agentic coding, with heavy RL on multi-step software work and Cursor data in supplemental training. Independent AA reads put it near the frontier with much better cost and token efficiency than many premium peers.
03What is Grok 4.5 best for?+
Fast agentic coding, terminal workflows, and knowledge work where cost and speed matter. It is especially strong in Grok Build style coding-agent setups, and competitive on Terminal-Bench and long-horizon coding while staying cheaper per finished task.
04What are some of the main limitations of this model?+
It is not the absolute top of every pure-coding board (Fable/Opus still lead some resolve-rate evals). AA also notes higher hallucination rate alongside higher knowledge accuracy. Harness choice matters: subscription vs API usage can differ across tools.
05What is a good example of using Grok 4.5 well?+
A bounded coding agent job: open a workspace, pick Grok 4.5, give it a repo task with tests (fix a bug, scaffold a feature, document the change), and let it run tools and terminal until green. That plays to its speed, cost, and agentic coding strengths.
06Where is Grok 4.5 best to test for full agentic capabilities?+
SPIRITT Workspaces. Pick Grok 4.5 in the model picker and run real work in a fully equipped cloud environment: tools, browser, files, terminal, memory, and durable sessions. Available on every plan, including trial. That is where the model can act as an agent, not just chat.
Buy from builders who use what they sellBuilt using