GPT-5.5-Codex-Spark Appears on Pricing Page

OpenAI’s latest coding model has surfaced on a pricing sheet rather than through an official announcement. The Pro plan description on the Codex pricing page now quietly lists 'GPT-5.5-Codex-Spark (research preview).' With no official blog post or model card published, the updated catalog copy offers the first confirmation of the new model's existence.

The manner of the disclosure is particularly telling. While the page's usage-limit table still references older versions, namely GPT-5.4 and GPT-5.3-Codex, the plan description has already been updated to include GPT-5.5, GPT-5.5 Pro, and GPT-5.5-Codex-Spark. Such premature updates to promotional copy ahead of system tables typically signal that a public rollout is imminent.

Spark: A Coding Product Line Engineered for Speed

To understand the significance of the addition, one must trace the origins of the 'Spark' designation. Spark is OpenAI’s ultra-low-latency product line, powered by Cerebras silicon and designed specifically for real-time coding applications. Its predecessor, GPT-5.3-Codex-Spark, debuted in February as a research preview in partnership with Cerebras Systems, leveraging Cerebras' wafer-scale hardware to deliver generation speeds exceeding 1,000 tokens per second alongside a 128k context window. The architecture is optimized for workflows where latency directly impacts developer productivity, such as real-time code modifications and interactive, multi-turn programming sessions.

The new iteration continues this lineage. The underlying base model, GPT-5.5, was released on April 23 as OpenAI's flagship offering, priced at $5 per million input tokens and $30 per million output tokens, with a 1-million-token context window via its API. Within the Codex ecosystem, most standard paid tiers currently operate with a 400k context window, while 'Fast Mode' generates tokens 1.5 times faster at 2.5 times the base cost. GPT-5.5-Codex-Spark appears to be a high-speed variant of this flagship model, optimized specifically for the Spark framework. Like its predecessor, it is expected to arrive as a research preview on Cerebras-backed infrastructure.

Backend Logs and Prediction Markets Point to GPT-5.6

The industry's competitive focus on speed shows no signs of abating. In mid-May, OpenAI's next-generation model briefly appeared in Codex backend logs. Operating under the codename 'iris-alpha,' this iris build reportedly features a 1.5-million-token context window, though OpenAI has yet to issue an official statement about the iris model. Further technical details were discussed in our earlier report.

Market participants have already turned their attention to June. On the prediction platform Polymarket, as of June 2, the implied probability of a June 8–14 launch stands at 68.5%, with bets heavily concentrated in that specific week. The cumulative probability rises to 77% by June 15 and reaches 93% by the end of the month. Traders are even interpreting a minor update to GPT-5.5 on May 28 as an indication of continuous, active development.

These two developments point to a busy June for OpenAI. While the pricing page prematurely exposed GPT-5.5-Codex-Spark, Polymarket and backend logs simultaneously point to a June release. Whether OpenAI intends to announce both models concurrently remains to be seen, but the pricing page updates and market sentiment strongly suggest that the company’s next significant move will occur this month.