Grok 4.5: xAI’s Fast, Low-Cost AI Model — What It Is and How to Chat With It Free

Grok 4.5 is the latest large language model from Elon Musk’s xAI, released on July 8, 2026. It is xAI’s smartest model to date, tuned for coding, agentic tasks, and everyday knowledge work — pitched not as the single most powerful model, but as the fastest and cheapest of the frontier tier.

Below you’ll find what Grok 4.5 actually is, its headline features and benchmarks, how it stacks up against Claude Opus and GPT-5, and how to start chatting with it free online in seconds.

What Is Grok 4.5?

Grok 4.5 is a large language model developed by xAI, publicly released July 8, 2026 after a late-June private beta at SpaceX and Tesla. It’s the latest step in the Grok line, which started with Grok 4 in mid-2025 and moved through Grok 4.3 and Grok Build 0.1 before this release. For background on the full model lineage, see the Grok chatbot entry on Wikipedia; the launch itself is documented in xAI’s official announcement.

The model and who built it

xAI, the company Musk founded to build Grok, positions 4.5 as its strongest model to date rather than a marginal update. The release sequence leading up to it:

• Grok 4 — released mid-2025
• Grok 4.3 — incremental update to the Grok 4 line
• Grok Build 0.1 — first coding-focused release
• Grok 4.5 — released July 8, 2026

The jump from Grok 4.3 to Grok Build 0.1 and now to 4.5 reflects a shift in focus: earlier versions targeted general chat and reasoning, while 4.5 is explicitly tuned toward coding and practical office work. That shift shows up throughout the model’s training data and default deployment, described below.

Built on the 1.5T V9 foundation

Grok 4.5 runs on xAI’s 1.5-trillion-parameter V9 foundation model, a mixture-of-experts architecture that marks a step up from earlier Grok versions. It was trained in xAI’s Memphis data centers on tens of thousands of NVIDIA GB300 GPUs — part of the Colossus supercomputing cluster — on datasets spanning coding, science, engineering, and mathematics. That scale of compute is one reason xAI can serve the model both quickly and cheaply, a combination explored further below.

What’s New: Key Features of Grok 4.5

The headline change in Grok 4.5 isn’t a single benchmark score — it’s where xAI aimed the model. Two areas stand out: coding, and broader “knowledge work” beyond a code editor.

Trained alongside Cursor for real coding

Grok 4.5 was trained alongside Cursor, the AI coding editor, incorporating real developer-session data — debugging traces and live code edits — rather than relying only on static repositories. This is why the model performs well on agentic, multi-file coding tasks: challenging Rust and C/C++ work, and end-to-end app building from a single prompt through to a working product.

Beyond code: office and knowledge work

Grok 4.5 is the default model inside Grok Build, and its scope extends well past code:

• Finance and legal document review
• Complex, multi-sheet Excel models built with live web research
• Native PowerPoint diagrams generated from a prompt
• Structured, long-form Word prose

xAI also ships Microsoft Office plugins for Word, PowerPoint, and Excel, putting the model directly inside everyday office workflows rather than requiring a separate chat window.

Grok 4.5 Benchmarks and Performance

Numbers matter more than marketing copy here, so it’s worth looking at where Grok 4.5 actually lands against its closest rivals before deciding whether the speed and price advantage below is worth the tradeoff.

Where it lands on coding benchmarks

On SWE-Bench Pro — a resolution-rate benchmark for real-world coding tasks — Grok 4.5 scores 64.7%, ahead of GPT-5.5 (58.6%) but behind Claude Opus 4.8 (69.2%) and Claude Fable 5 (80.4%). On DeepSWE, xAI’s own published launch chart lists two revisions side by side: Grok 4.5 scores 62.0% on the DeepSWE 1.0 revision but 53% on the newer, harder DeepSWE 1.1 revision (where Claude Opus 4.8 scores 59% and GPT-5.5 scores 67%) — independent coverage confirms both figures came from xAI’s own data, so the two shouldn’t be averaged into a single score.

The bottom line: Grok 4.5 is competitive on coding benchmarks, not category-leading. Its case rests elsewhere.

Fast, and unusually token-efficient

Grok 4.5 is served at 80 tokens per second — fast-model territory — and roughly doubles the token efficiency of comparable frontier models. On SWE-Bench Pro tasks it used around 15,954 output tokens on average, versus roughly 67,020 for Claude Opus 4.8 — a 4.2x gap that compounds into meaningful savings at high volume.

Grok 4.5 vs Claude Opus and GPT-5

xAI itself has framed Grok 4.5 as sitting close to Claude Opus in raw capability, even though the benchmark table above tells a more nuanced story. It’s worth separating the marketing claim from the measured results.

Musk’s “Opus-class” positioning

Grok 4.5 is an Opus-class model, but faster, more token-efficient and lower cost. Roughly comparable to Opus 4.7, but much faster.

Elon Musk, via TechCrunch

That’s xAI and Musk’s own framing of the model, not an independent benchmark verdict — the SWE-Bench Pro numbers above show Opus 4.8 and Fable 5 both outscoring Grok 4.5 on coding specifically. Where Grok 4.5 does hold up is speed, token efficiency, and cost, which is where the comparison gets more interesting.

The real argument is price

The strongest case for Grok 4.5 is cost. At $2 per million input tokens and $6 per million output tokens, it undercuts both major rivals by a wide margin:

• Grok 4.5 — $2 input / $6 output per million tokens
• Claude Opus 4.8 — $5 input / $25 output per million tokens
• GPT-5.6 Sol — $5 input / $30 output per million tokens

Combined with the token-efficiency edge described earlier, teams running high request volumes can iterate for a fraction of the cost while getting roughly Opus-4.7-class capability for most everyday tasks.

Pricing and Context Window

Beyond the headline comparison, two practical numbers matter most for anyone deciding whether to build on Grok 4.5: per-token pricing and context length.

API pricing at a glance

At $2 per million input tokens and $6 per million output, Grok 4.5 runs at roughly half the per-token price of comparable flagship models, even before accounting for its efficiency gains. Prompt caching can lower the effective price further on gateways such as OpenRouter, which also lists live pricing and benchmark data for the model.

500,000-token context

Grok 4.5 offers a 500,000-token context window — about 400,000 words — which is enough to hold large codebases or long documents in a single request without chunking.

How to Chat With Grok 4.5 Free Online

Getting access to Grok 4.5 doesn’t require picking a single plan — there are several free and paid entry points depending on how you plan to use it.

Free and paid entry points

You can chat with Grok 4.5 free online right here on this page — no signup required for a quick test. Beyond that, access breaks down like this:

1. Free, limited-time access — xAI is offering free Grok 4.5 usage for a limited time inside Grok Build and Cursor.
2. Subscription access — X Premium and SuperGrok unlock Grok 4.5 inside the Grok interface on X.
3. Developer access — the xAI API (model id grok-4.5) via api.x.ai, or through gateways like OpenRouter, for teams building their own applications.
4. Editor integrations — Cursor offers Grok 4.5 across all its plans, and Warp includes it for terminal-based agentic workflows.

Pick the entry point that matches what you’re trying to do: casual chat, coding inside an editor, or programmatic API access.

Regional availability

One gap worth noting: at launch, Grok 4.5 was not available in the EU through xAI’s own products or the API console, with availability expected to open up in mid-July 2026. Users outside the EU can access the model immediately through any of the channels above.

Learn More About Grok 4.5

Explore our full, fact-checked Grok 4.5 reference:

• Grok 4.5 model overview — capabilities, benchmarks and what’s new
• Grok 4.5 API guide — access, pricing and a quick start
• How to use Grok 4.5 — a complete guide across app, web and API
• Grok 4.5 use cases — real examples of what Grok can do