Warp: GPT Models Unlock 1M Context Window for Deep Codebase Reasoning
WarpWarp added configurable 1M context window support for GPT-5.4 and GPT-5.5 (and the grape model alias) in its June 3, 2026 stable release. Developers using these models in Agent Mode can now instruct Warp agents to reason across much larger codebases and conversation histories without hitting context limits. Warp also surfaces a pricing warning when context usage exceeds OpenAI's 272K threshold, keeping users informed before costs scale.
Sources & Mentions
4 external resources covering this update
GPT Models Get 1M Context Support in Warp
Warp's June 3, 2026 stable release expands the context ceiling for OpenAI's newest models inside the Warp agent environment. GPT-5.4, GPT-5.5, and the grape model variant now support a configurable 1M token context window, up from the previous default of 128K. This allows agents running in Warp to hold significantly more of a project's codebase, shell history, and conversation context in memory at once.
What Changed
The context window for supported GPT models is now configurable per profile in Warp's Agent Mode. The server acts as the source of truth for which models support expanded context, so the option only appears when the underlying model is capable of it. For models that do not support configurable context windows, the setting remains hidden, keeping the interface clean.
A visible pricing caution was added alongside the expanded context option. When an agent session's context usage climbs beyond OpenAI's 272K input token threshold, Warp displays a warning indicating that OpenAI applies a 2x input cost and 1.5x output cost multiplier for the full session at that scale. This is a meaningful quality-of-life addition: large-context sessions with GPT-5.5 can become significantly more expensive than standard sessions, and the warning gives developers an opportunity to trim context before costs compound.
Why It Matters
GPT-5.5's 1M context window was already available through the OpenAI API, and Warp's partnership with OpenAI, which includes OpenAI as founding sponsor of Warp's open-source repository, means Warp users benefit from access to frontier model capabilities early. Unlocking the full context ceiling inside Warp means agents can now tackle tasks that require sustained awareness of large, multi-file codebases: long refactors, cross-module migrations, or architectural analysis that would previously require manual context management.
The pricing warning reflects a broader design philosophy: giving developers visibility into cost implications directly in the tool, rather than surfacing billing surprises after the fact.
Also in This Release
The June 3 release rounded out several stability and polish items: a fix for cannot-detect-diffs errors, improvements to narrow cloud mode input layout, a fix for secret redaction after multibyte UTF-8 prefixes, and resolution of queued prompt edge cases introduced in the prior release.