Three Codenames, One Upcoming Model
As OpenAI moves toward launching GPT-5.6 — expected in June 2026 — three internal codenames have leaked from developer logs and researcher reports: ember-alpha, beacon-alpha, and iris-alpha. None of these are official OpenAI designations, and the company has not confirmed them. But the pattern is consistent with how frontier labs name experimental builds ahead of public release.
What ember-alpha Likely Represents
The “ember” codename has appeared most frequently in Codex-related log traces. Researchers tracking the leak believe ember-alpha is the primary experimental checkpoint — the build that briefly appeared as a Codex routing entry in mid-May 2026 before vanishing from subsequent session files. The “alpha” suffix is standard for pre-release builds that are runnable but not yet production-ready. ember-alpha is likely the core GPT-5.6 model under active canary testing with real Codex traffic.
What beacon-alpha Likely Represents
beacon-alpha has been referenced alongside ember-alpha in multiple independent reports. The “beacon” name suggests a model variant optimised for signal clarity — potentially a version tuned for lower latency and faster token throughput. This aligns with rumours of a GPT-5.6 “UltraFast Codex mode” designed to compete directly with Claude Code on response speed for agentic software tasks.
What iris-alpha Likely Represents
iris-alpha is the least-cited of the three names but has appeared in social posts from researchers monitoring OpenAI’s API infrastructure. “Iris” — referencing vision — may indicate a multimodal variant of GPT-5.6 with enhanced image and document reasoning capabilities. If accurate, this would represent a continuation of OpenAI’s push to make vision-capable models available across all tiers of ChatGPT.
What the Alpha Designation Means
All three codenames carry the “alpha” suffix, which in standard software development signals a pre-release build that exists and is functional but is not yet cleared for general availability. The presence of multiple alpha builds — rather than a single one — suggests OpenAI is running parallel capability tests, likely comparing different fine-tuning approaches or architectural configurations before selecting the final GPT-5.6 release configuration.
What Is Actually Confirmed
As of June 16, 2026, OpenAI has made no official announcement about GPT-5.6. What is confirmed: a rollout-mapping entry referencing gpt-5.6 appeared briefly in OpenAI’s Codex backend logs before disappearing. Prediction markets on Polymarket and Manifold are pricing an 80–89% probability of a public GPT-5.6 release by June 30, 2026. The codenames add colour to the leak story but do not constitute an announcement or a confirmed feature list.
For full context on the GPT-5.6 release timeline, see our GPT-5.6 release news post and our GPT-5.6 features and pricing guide.