Slipstream already works with GPT-6 Astra.

Why nothing had to change

Slipstream doesn't talk to OpenAI, Anthropic, or any model provider directly. It wraps your agent CLI — Codex, Claude Code, Hermes — and points it at a local proxy that sits between your agent and whichever provider you've already configured. The proxy compresses context on the way out; it doesn't care what model answers on the other end. If your Codex install is set to GPT-6 Astra, that's what it talks to, exactly the same way it talked to GPT-5 yesterday.

Practically: if you're already on Slipstream, you don't need to update, reconnect, or reconfigure anything to use GPT-6 Astra through it.

The one honest caveat

Compression works on any model. The dollar-savings number is a different matter — it depends on headroom, the compression engine Slipstream bundles, knowing the right tokenizer and price-per-token for the model you're running. Headroom's current release predates Astra's launch by about a week, so it doesn't yet ship an Astra-specific entry. Until it does, Slipstream's estimate for Astra traffic falls back to an approximation rather than an exact figure — the compression itself is unaffected, just the number on the dashboard.

Improve it yourself, today

You don't have to wait on us or on headroom. Settings → Models lets you add any model by name with its context limit and price-per-token — it's a plain editable list, not a fixed dropdown. Add an entry for GPT-6 Astra with OpenAI's published pricing and Slipstream will use it immediately for that model's savings math. We'll ship an accurate default the moment headroom's own tokenizer support lands.

Why bigger context makes compression matter more

A 1M-token context window doesn't mean your agent should fill it. Every stale file, repeated tool result, and dead-end exploration that stays in context gets re-read, and re-billed, on every following turn — that cost scales with the window you have available, not just the one you need. Bigger context windows raise the ceiling on how much waste can quietly accumulate before anyone notices. Slipstream's job stays the same regardless of which model is on the other end: keep what your agent needs in context and cut the rest before it's billed.