Grok 4.5 on Autohive now gets cheaper in long conversations

xAI cut Grok 4.5’s cached input pricing by 40% on 28 July 2026. The cut covers both context tiers, and if you run Grok 4.5 through longer, repeated conversations on Autohive, you pay less for that repeated context without changing a single setting.

What cached input actually is

Every time a model processes a conversation turn, it reads the whole context: the system prompt, the conversation history so far, any attached documents, and the new message.

In a longer conversation, most of that context is not new. The model still receives the earlier turns, the same instructions, and any context already attached to the conversation. That repeated conversation context is where cached input pricing matters.

Cached input pricing is the discount a provider applies when it recognizes it has already processed a block of tokens. Instead of billing the full input rate again, it bills a lower cached rate for the repeated portion. xAI covers the Grok 4.5 caching behavior in its developer docs.

For Grok 4.5, which supports a 500k token context window, that can add up quickly in longer sessions. A short, single-question exchange will not see much difference. A deeper agent conversation that keeps building on the same context will.

The new pricing

xAI cut cached input pricing by 40% across both context tiers on 28 July 2026. Regular input and output rates did not change. Full current rates are in the xAI pricing docs.

Where this shows up on Autohive

The clearest benefit is in longer, multi-turn conversations with Grok 4.5.

Every conversation with an Autohive agent adds to its own context as it goes. By turn five or six, much of what gets sent to the model is conversation history rather than brand-new input. That repeated conversation context is now cheaper at the cached rate.

If the conversation includes workspace content, uploaded files, or other attached context, that content can also be part of the repeated context inside the same conversation. The important point is that this benefit applies inside the ongoing conversation. It should not be read as a saving that automatically carries across separate runs.

No workflow changes needed for longer conversations

If you are already running Grok 4.5 on Autohive and your longer conversations carry repeated context, the lower cached-input price applies where xAI caching applies. There is nothing to turn on or configure in Autohive for this pricing change.

The saving depends on how much repeated context a session carries. A quick one-turn task has little repeated context. A longer agent conversation, especially one working through several steps with the same background context, is where the change matters.

Assign Grok 4.5 to your agents

Grok 4.5 is available on Autohive through direct xAI integration. Assign it to any agent from agent settings.

If you are running agents on other models and want help deciding which ones would gain the most from switching, this post on per-agent model routing walks through the decision.

For the full spec on Grok 4.5, including the 500k context window, throughput, and capability details, check the xAI developer docs.