Grok 4.5 Arrives With Faster Agentic Reasoning

xAI released Grok 4.5 on July 16 with a straightforward pitch: keep frontier reasoning useful while making the model fast enough for interactive coding, research, and agents.

The model is now available to Poly subscribers through OpenRouter. It brings text and image understanding, file input, tool use, configurable reasoning, and a 500,000-token context window into the same workspace as the rest of Poly's model lineup.

Speed Is the Product Change

xAI reports output speeds around 80 tokens per second for Grok 4.5. Benchmarks can help compare releases, but that throughput is the part users will notice in daily work: plans appear sooner, code iterations feel more conversational, and a tool-using agent spends less time waiting between steps.

That matters most when a model is not producing one answer. An agent may inspect a file, decide on a tool, read the result, revise its plan, and repeat. Lower latency at every turn compounds across the whole run.

Where Grok 4.5 Fits

Grok 4.5 is a strong candidate for:

• Coding sessions with frequent back-and-forth edits
• Research that mixes large documents with current web context
• Tool-using agents that need a responsive reasoning loop
• Long-form synthesis across many files
• Teams that want a different model's second opinion on a hard problem

Its 500,000-token context window is smaller than the one-million-token windows appearing in several other 2026 releases, but it is still enough for substantial repositories and document collections. More context is not automatically better; the useful measure is whether the model can find and use the right details inside it.

Multimodal Inputs and Tools

Through OpenRouter, Grok 4.5 accepts text, images, and files. It also supports tool calls and reasoning controls. You can ask it to inspect a screenshot, connect that evidence to a specification, then use a tool to act on the result.

This is the shift behind many current model releases. The model is no longer only a response generator. It is the decision layer inside a workflow.

Price and Access

OpenRouter currently lists Grok 4.5 at $2 per million input tokens and $6 per million output tokens. That places it below the premium pricing of the largest flagship models while keeping the features needed for serious agent work.

On Poly, Grok 4.5 is a subscriber model. The model picker shows its context window, capabilities, and routed price alongside alternatives, so you can choose it for the tasks where speed changes the experience.

When to Choose It

Choose Grok 4.5 when you want a fast agentic model with a large context window and strong coding focus. For a final answer that is expensive to get wrong, compare its work with another frontier model. Multi-model review remains one of the simplest ways to catch blind spots.

Try Grok 4.5 on Poly Compare Plans

Source: xAI's Grok 4.5 announcement.