OpenAI Cuts Codex's Context Threshold

OpenAI has lowered the context threshold that triggers premium billing in Codex, its coding agent built on GPT-5.6. The effective cutoff dropped from 372,000 tokens to 272,000 tokens, a cut of roughly 27%. That means sessions shorter than before can now trigger higher rates, a shift many developers who still plan around the old number may not have noticed.

How the Premium Pricing Above 272K Tokens Works

Once a single session crosses 272,000 input tokens, OpenAI bills the entire session at a premium rate: double the normal input token price and 1.5 times the normal output token price. This is not a hard technical cap, the model can still process more tokens than that. It is a billing tier boundary: crossing 272K tokens changes the price for the whole session, not just for the tokens above the line.

Multiple Independent Pricing Analyses Confirm the Same Figures

The 272K threshold and the 2x input, 1.5x output multiplier show up consistently across independent pricing guides from outlets specialized in AI API costs, including AIWeekly, CloudZero, Eesel AI, Verdent, Layer3Labs, and AnyCap. That convergence across separate sources, rather than a single official OpenAI announcement, is what gives the figures credibility: it reflects a market wide reading of how the new pricing scheme behaves in practice.

The Bigger Picture: Token Based Billing Since April

The change fits into a broader shift OpenAI has been rolling out in Codex since around April 2026, when the product moved from per task or per message billing to a credit system based on actual token consumption. According to the same pricing guides, that transition has made real costs harder to predict for developers: a complex, multi file coding task can reportedly consume up to roughly nine times the credits of a simple, single file script edit.

The Risk for Integrations Built Around the Old Limit

Any Codex integration, custom agent harness, or automation workflow still designed around the old 372K threshold is one long context session away from an unexpected bill at the 2x and 1.5x premium rates. Because the new threshold is lower, it is easier to cross without noticing, especially in tasks that accumulate conversation history, large files, or multiple iterations within the same session.

Why It Matters for Brazilian Agencies and SMBs

Brazilian automation agencies that rely on Codex or similar coding agents to build and maintain their own internal tooling need to watch context length in their workflows and prompts to avoid surprise bills at the premium multiplier. It is also a broader reminder: teams managing AI integrations for clients should closely monitor token based billing terms across every vendor used in client projects, since pricing thresholds can shift with little notice and directly hit margins on fixed scope engagements.