Claude 1M Goes GA with No Price Increase – Why Claude Code Users Are Excited

The 1‑million‑token context window for Claude is now generally available (GA) and the pricing for inputs and outputs above 200K tokens has been removed, so the cost remains the same as the previous 200K tier.

Feature changes: The default context window is now 1M tokens (previously 200K default, 1M required manual activation). No surcharge for exceeding 200K input or output tokens.

Pricing comparison (per million tokens):

Claude Sonnet 4.6 – $3 base, no overage, total $3.00

Gemini 3.1 Pro – $2 base, over‑200K usage doubles price, total $3.60

GPT‑5.4 – $2.50 base, over‑272K usage doubles price, total $3.95

Claude Opus 4.6 – $5 base, no overage, total $5.00

When the full 1M token limit is utilized, Claude Sonnet becomes the cheapest option, 17% cheaper than Gemini and 24% cheaper than GPT.

Why Claude Code users should be excited: With 1M tokens the entire codebase (50‑80 k lines, ~0.5‑0.8 M tokens) can be fed to the model at once, eliminating the need to select individual files. This enables:

Cross‑file refactoring without fragmented context.

One‑pass bug tracing across the whole call chain.

Instant project‑level architecture overviews for new team members.

Real‑world case: An Express + React full‑stack project that required three iteration cycles with a 200K window because the model missed a backend middleware dependency. Switching to the 1M window allowed the entire front‑ and back‑end code to be processed in a single pass, achieving interface renaming and front‑end adaptation with zero errors.

Model selection guidance:

For project‑scale analysis, choose Claude Sonnet 4.6 for the best cost‑performance.

Cursor or Windsurf users should switch their model to Claude Sonnet 4.6 for more predictable large‑repo costs.

For small functions or short conversations, any of the three models (Claude, Gemini, GPT) perform similarly.

Impact on Retrieval‑Augmented Generation (RAG): Previously, a 200K limit forced many users to split documents, store embeddings, and stitch relevant chunks into the prompt. With a 1M window, scenarios such as codebases under 800 k tokens, product/API documentation under 500 k tokens, or one‑off analyses (contract review, competitor reports) can skip RAG entirely. However, massive knowledge bases (thousands of documents) and real‑time data streams still require RAG.

Three common pitfalls to avoid:

Do not assume the 1M limit should always be filled; using the full window slows responses and raises costs.

All large models suffer from “lost in the middle” – middle sections may be forgotten, so place critical information at the beginning or end.

Haiku 4.5 remains limited to 200K tokens; only Opus and Sonnet support the million‑token context.

Action checklist:

Set Claude Code model to claude-sonnet-4-6 or claude-opus-4-6 – 1M context is automatic.

In Cursor/Windsurf, switch the model setting to Claude Sonnet 4.6 for immediate upgrade.

When calling the API, keep the model parameter unchanged; the context size expands to 1M automatically.

If you have a simple RAG pipeline and the total token count is under 1M, remove the RAG layer to save maintenance effort.

Moving from 200K to 1M tokens is not just a five‑fold increase in capacity; it transforms AI from “can only see a few files” to “can see the entire project” without extra cost.