Brief IA

OpenAI Revolutionizes Codex with Token-Based Pricing

🛠️ AI Tools·Tom Levy·

OpenAI Revolutionizes Codex with Token-Based Pricing

OpenAI Revolutionizes Codex with Token-Based Pricing
Key Takeaways
1As of April 2, 2026, OpenAI charges for Codex by token, replacing the previous message-based system.
2The costs for GPT-5.3-Codex are $1.75 per million tokens for input and $14 per million for output.
3This change aims to better reflect actual consumption, impacting heavy users with potentially higher costs.
💡Why it mattersThis new pricing could increase costs for heavy developers while aligning OpenAI with competitors like Anthropic.
Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

OpenAI Reinvents Codex Pricing

Since April 2, 2026, OpenAI has introduced a new billing method for Codex, its AI agent specialized in code. This change aims to replace the old message-based system with token-based billing, already in place for the API. This transition begins with Business and Enterprise clients and is expected to soon extend to Plus and Pro subscribers.

Credits remain the basic unit for billing, but their calculation is now directly linked to token consumption. Credits are deducted based on three types of tokens:

  • Input tokens
  • Cached input tokens
  • Output tokens

For the GPT-5.3-Codex model, the rate is set at $1.75 per million input tokens and $14 per million output tokens. The context window capacity reaches up to 400,000 tokens, and the fast mode doubles credit consumption. OpenAI estimates that the average monthly cost per developer is between $100 and $200, although this varies depending on the model used and the intensity of usage.

Increased Transparency or Rising Costs?

OpenAI presents this new pricing as a way to improve cost transparency, allowing for a better match between actual consumption and billed credits. The old message-based system did not always reflect the reality of resources consumed, charging the same for tasks of varying complexity.

However, this change could also lead to increased costs for heavy users. Those who use Codex for complex and frequent tasks, especially in fast mode, may see their bills rise significantly. In contrast, for developers performing lighter tasks, the impact should be limited.

A Strategy Aligned with Anthropic

OpenAI appears to be shifting towards a strategy focused on professional clients, who already account for over 40% of its revenue. The goal is to achieve parity with the consumer market by the end of 2026. This approach is reminiscent of Anthropic, which has successfully established itself with its product Claude Code.

The strategy seems to be paying off: Codex has surpassed 3 million weekly users, up from one million at the beginning of the year. However, OpenAI is looking to improve Codex's profitability, as its financial situation remains precarious. For the first time, Anthropic announced an annualized revenue exceeding that of OpenAI, surpassing $30 billion after tripling its revenue in a single quarter. By adopting a pricing model based on actual consumption, OpenAI hopes to better monetize intensive usage rather than subsidizing it through a fixed fee.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.