What Is Grok 4.5? SpaceXAI Frontier Model Specs
Grok 4.5 is SpaceXAI's first major release since going public. It was co-trained with Cursor after SpaceX acquired Anysphere in June 2026, using trillions of tokens from real IDE sessions — code review, debugging, and agent-to-codebase interactions.
Built for:
Coding agents: Bug fixes, large refactors, end-to-end app builds.
Agentic automation: Multi-step workflows across enterprise apps.
Knowledge work: Legal, healthcare, education, data analysis.
High-volume cost pressure: Teams running hundreds of agent calls daily feel API bills immediately.
First-shot accuracy: Complex stateful UI tasks punish models that need retries.
| Spec | Detail |
|---|---|
| Architecture | Mixture of Experts (MoE) |
| Context window | 500,000 tokens |
| Reasoning modes | Low / Medium / High (default: High) |
| Speed | 80 TPS official, ~90 TPS measured |
| Training infra | Tens of thousands of NVIDIA GB300 GPUs (Memphis, TN) |
| Parameters | Not disclosed |
Grok 4.5 Pricing: How Much Can You Actually Save?
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Grok 4.5 | $2.00 | $6.00 |
| Grok 4.5 (cached input) | $0.50 | — |
| Grok 4.5 Fast | $4.00 | $18.00 |
| Claude Opus 4.7 | $5.00 | $25.00 |
| GPT-5.6 Sol | $5.00 | $30.00 |
| GPT-5.6 Luna | $1.00 | $6.00 |
| Platform | Avg tokens/task | Cost per task |
|---|---|---|
| Grok 4.5 / Grok Build | ~1.9M | $2.49 |
| GPT-5.5 / Codex | ~6.2M | $5.07 |
| Claude Fable 5 / Claude Code | ~7.2M | $11.80 |
On SWE-Bench Pro, Grok 4.5 averaged 15,954 output tokens per task vs 67,020 for Claude Opus 4.8 — a 4.2× efficiency gap. At 500 tasks/day, that is roughly $1,245/day vs $5,900/day.
Grok 4.5 Benchmark Results: Coding vs Agentic Work
| Benchmark | Grok 4.5 | Fable 5 | Opus 4.8 | GPT-5.5 |
|---|---|---|---|---|
| DeepSWE 1.0 | 62.0% | 66.1% | 55.75% | 64.31% |
| DeepSWE 1.1 (neutral) | 53% | 70% | 59% | 67% |
| Terminal Bench 2.1 | 83.3% | 84.3% | 78.9% | 83.4% |
| SWE-Bench Pro | 64.7% | 80.4% | 69.2% | 58.6% |
Caveat: CursorBench was pulled after a Cursor codebase snapshot contaminated training data — a transparency problem with the launch.
| Agent benchmark | Grok 4.5 | Fable 5 | Opus 4.8 |
|---|---|---|---|
| AutomationBench-AA (657 tasks) | 51.4% | 48.6% | 48.5% |
| Snorkel GDPVal+ | 29% | — | 21% |
Grok 4.5 is the first model to complete more than half of AutomationBench-AA workflow objectives without violating business constraints. Snorkel shows wide leads in legal (40% vs 27–28%), education (58% vs 35–42%), and healthcare (35% vs 23–25%). Artificial Analysis Intelligence Index: 54/100 (fourth), up 16 points over the previous Grok.
TryAI real coding tests: Opus 4.8 and Fable 5 passed the 3D cube on the first try; Grok 4.5 needed a retry; GPT-5.5 failed. Grok delivered first token in under 500ms at ~110 tokens/sec — roughly twice as fast as competitors.
How to Use Grok 4.5 in Cursor and the API
Available on Grok Build, all Cursor plans (desktop, web, iOS, CLI, SDK), SpaceXAI Console API, Microsoft Office add-ins, and gateways (OpenRouter, Vercel, Cloudflare, Snowflake, Databricks Mosaic). Regions: us-east-1, us-west-2. Limits: 150 req/s, 50M tokens/min. EU expected mid-July 2026.
curl -s https://api.x.ai/v1/responses \
-H "Authorization: Bearer $XAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "grok-4.5",
"input": "Find and fix the bug: function median(a){a.sort();return a[a.length/2]}"
}'
Create an API key in SpaceXAI Console and confirm US region access.
Select Grok 4.5 in Cursor from the model picker; launch-week usage was doubled.
Set prompt_cache_key (Responses API) or x-grok-conv-id header to cut cached input to $0.50/M.
Enable Context Compaction for long agent loops.
Route routine tasks to Grok 4.5 and escalate architecture decisions to Claude Fable 5.
Validate outputs in production; hallucination rate on AA-Omniscience hit 54%.
Should You Switch? Verdict and Hard Numbers
Grok 4.5 makes sense for high-volume agentic pipelines, terminal-heavy workflows, Cursor-native teams, cost-sensitive startups, and mixed-model routing. Think twice for SWE-Bench Pro-grade precision, hallucination-sensitive systems, EU deployments before mid-July, and CursorBench-trusted claims.
Per-task cost: ~$2.49 (Grok) vs ~$11.80 (Claude Code).
Output token efficiency: 15,954 vs 67,020 (4.2× gap vs Opus 4.8).
Throughput: 80 TPS official, ~90–110 tokens/sec measured, first token <500ms.
Grok 4.5 is not the most accurate coding model — Claude Fable 5 holds that crown. But accuracy per benchmark is not the same as value per dollar. API-only setups work for trials; local Macs often hit memory pressure under long agent loops, and VMs sacrifice Metal/Xcode fidelity. For 7×24 Cursor agents, parallel dev servers, and iOS CI/CD, MESHLAUNCH cloud Mac Mini bare-metal rental is usually the better production fit: dedicated Apple Silicon and flexible daily/weekly/monthly terms.
Sources: x.ai · Cursor · API docs · TechCrunch · Snorkel AI
Depends on the metric. Opus wins SWE-Bench Pro accuracy; Grok wins speed, token efficiency, and per-task cost — often by 4x. See our pricing page to budget dev hardware.
Limited free usage was offered in Grok Build and Cursor at launch. Ongoing API pricing is $2/M input and $6/M output.
Pick Grok 4.5 from the model selector on any Cursor plan. Check the help center for deployment questions.
Training data contamination from a Cursor codebase snapshot invalidated those numbers.
Yes — also on Vercel AI Gateway, Cloudflare, Snowflake, and Databricks Mosaic.