Grok 4.5
xAI's Grok 4.5 — the coding/agentic model, default in Grok Build. Tops SWE Marathon, ~4x more token-efficient than Opus 4.8, ~80 TPS.
Reference benchmarks for Grok 4.5
These are external benchmarks I pulled from the source comparison guides on agentos.guide — SWE-bench Verified, DRACO, Kilo plan rubric, build-time measurements, vendor-reported coding scores. They are not goldiebench medal scores (those come only from same-prompt one-shot creative coding tasks in the matrix). I surface them here so the spec sheet for Grok 4.5 is honest about what's measured.
What is Grok 4.5?
Grok 4.5 is the xAI · Grok Build frontier model with a xAI's smartest model, built for coding + agentic tasks; trained alongside Cursor. Default model in Grok Build. Benched here via OpenRouter (x-ai/grok-4.5). context window, released 2026-07-08. Tagline: xAI's Grok 4.5 — the coding/agentic model, default in Grok Build. Tops SWE Marathon, ~4x more token-efficient than Opus 4.8, ~80 TPS.. Official source: x.ai/cli (Grok Build).
Pricing detail. ~4.2x fewer output tokens than Opus 4.8 on SWE-Bench Pro (15,954 vs 67,020 avg) and served at ~80 TPS, so real cost/latency is well below the sticker. Free for a limited time in Grok Build + Cursor.
How I use it inside the Agent OS. Benched one-shot on the same GoldieBench game prompts as the field, with the threejs-game-director patterns baked into each prompt; weak builds iterated by Grok 4.5 itself (the model authors every fix, never hand-patched). Wired into the Agent OS as the newest engine.
What I built with Grok 4.5
Every model on Goldie Bench gets the same fixed prompt set — one shot, single HTML file out — and I score the result 0–10 inside the Agent Operating System. Here's what Grok 4.5 shipped on the bench: 50 one-shot demos across xAI's smartest model, built for coding + agentic tasks; trained alongside Cursor. Default model in Grok Build. Benched here via OpenRouter (x-ai/grok-4.5). of context. Of those, 48 are scored against the field with my honest 0–10 from the source guides at agentos.guide.
Strengths
- SWE Marathon resolution #1: 29.0% (Opus 4.8 26.0, Fable 24.0)
- ~4.2x more token-efficient than Opus 4.8 on SWE-Bench Pro (15,954 vs 67,020 avg output tokens); ~80 TPS
- Strong one-shot game builds: gorgeous multi-part heroes + layered worlds + cohesive HUDs first try (dragonrealm, crypt, skyrim)
Trade-offs
- Raycaster/FPS (doom) under-renders + walks out of bounds one-shot; needed multiple self-fix passes
- Occasional TDZ/init bug blanks a build to black (racing) — recovered by the model itself in one pass
- Not available in the EU until mid-July 2026
Best for
- One-shot end-to-end app + game builds from a single prompt
- Cost/latency-sensitive agentic coding loops (token-efficient + fast)
- Office-work automation (Excel/PowerPoint/Word via Grok Build)
Every demo by Grok 4.5
50 live demos, sorted by category. Click any tile to play the actual one-shot result. Verdicts and 0–10 scores are pulled from the source guides where I posted them publicly.
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVECompare Grok 4.5 against every other model
Every head-to-head featuring Grok 4.5. Verdicts shown for scored pairs.
See all 66 comparisons across every model →
Quick pill index
Direct comparisons against every other scored model on the bench:
Grok 4.5 vs Fusion Grok 4.5 vs Hermes MoA Grok 4.5 vs Claude Fable 5 Grok 4.5 vs Grok (X real-time) Grok 4.5 vs MiniMax M3 Grok 4.5 vs Fugu Ultra Grok 4.5 vs GLM-5.2 Grok 4.5 vs Fugu Mini Grok 4.5 vs Opus 4.8 Grok 4.5 vs Kimi K2.7 Grok 4.5 vs Qwable 5 27B Coder Grok 4.5 vs Claude Sonnet 5 Grok 4.5 vs Qwen 3.7 Grok 4.5 vs Agents-A1 Grok 4.5 vs Gemma 4 12B · MLX Grok 4.5 vs Laguna XS 2.1 Grok 4.5 vs Qwythos 9B Grok 4.5 vs LongCat-2.0 Grok 4.5 vs Hy3 Grok 4.5 vs Gemma-4 12B CoderRead more on agentos.guide:
Run this stack yourself.
Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 4,000+ founders shipping with it every day all live inside the AI Profit Boardroom.