GPT-5.6 release tracker: iris-alpha, the Codex leak, and what's actually known (June 2026)

GPT-5.6 is OpenAI's next minor version, expected mid-to-late June 2026 based on leaked codenames, Codex backend logs, and Polymarket odds. It has not been officially announced as of June 6, 2026. This is the tracker — what's verified vs speculation, the codenames that have surfaced, and what builders should plan around regardless of timing.

What's verified

What's not verified

• Architecture changes. A canary name is not an architecture spec. The gap from GPT-5.5 (released April 23, 2026) to GPT-5.6 is three weeks — not enough for a foundation rebuild.
• Pricing. GPT-5.5 just stabilised at $1.25/$10 per MTok. OpenAI rarely re-prices on a minor version; expect $5-15/$15-30 territory if GPT-5.6 lands in the standard production lineup.
• Benchmark numbers. Nothing has leaked. Any quoted score is a guess.
• Exact release date. Polymarket's 89% June-30 anchor is a crowd estimate, not an OpenAI commitment.

The release-cadence signal

OpenAI's release cycle has been compressing:

• GPT-5 → GPT-5.1: ~5 months
• GPT-5.1 → GPT-5.2: ~3 months
• GPT-5.2 → GPT-5.3: ~2 months
• GPT-5.3 → GPT-5.4: ~6 weeks
• GPT-5.4 → GPT-5.5: ~5 weeks
• GPT-5.5 → GPT-5.6: ~6-7 weeks projected

The cadence alone makes a mid-June release plausible. The Codex log adds independent confirmation. The Polymarket consensus adds market validation. None of those are a date commitment.

The realistic window

June 15 – July 5, 2026 is the band that fits all observable signals:

• Codex log appeared May 1 → typical preview-to-launch lag of 4-8 weeks
• Polymarket median resolution date: June 30
• Compressed cadence: ~6-7 weeks since GPT-5.5

Plan around the window, not the date.

What probably ships

Based on the pattern of every prior minor version + the iris-alpha metadata that has leaked:

• A 1.5M-token context window — if the routing log metadata is correct, this would be the largest context tier on the standard OpenAI lineup (vs 200K for GPT-5.5).
• Stronger agentic performance — every minor since 5.2 has prioritised tool-use reliability.
• Same pricing or modest increase — OpenAI has held GPT-5.5 list at $1.25/$10 for input/output; expect similar or +20-50%.
• Available immediately via Codex — same launch surface as every prior version.
• No new SDK — wire format unchanged. Existing OpenAI SDK code + ChatCompletions endpoint will work day-of.

What to do before launch

Whether you're on the OpenAI Responses API, the standard ChatCompletions endpoint, or a wrapper like Cursor / Antigravity, the prep is the same:

1. Define your eval before the model lands

The temptation on launch day is to flip the model name and ship. Don't — you lose the ability to A/B compare. Pick 20-50 tasks representative of your production workload and run them against GPT-5.5 NOW. When GPT-5.6 ships, the same eval gives you a clean delta.

2. Test the cost shape, not just the headline price

The headline-price delta of a minor version is usually small. The hidden delta is what the model writes — output token count. GPT-5.4 → 5.5 saw output verbosity drop ~15% on the same prompts, so even at flat pricing the bill changed. Run your eval and measure output token counts, not just input → output ratios.

3. Build on a wire format that supports both day-zero

OpenAI SDK + OpenAI-compatible endpoint covers most cases. If you're using an LLM gateway, verify day-0 support is committed. Anvat (the discount gateway we build) typically lands new OpenAI models within hours of launch via the unchanged /v1/chat/completions path.

Why this matters more than usual

The next 4-8 weeks are unusually dense for the frontier:

Three frontier launches in a six-week window. Whichever lab ships first sets the benchmark conversation for the rest of summer.

What we'll update here

This post is a tracker. We'll update it as new signals confirm:

• Specific model identifier appearing in /v1/models API responses
• A system_card or safety page on openai.com
• Codex 2.x.x release notes that mention the new model
• First independent benchmark numbers from Artificial Analysis or similar
• Official pricing page entry
• Polymarket resolution

Related coverage

• Claude Mythos & Project Glasswing — Anthropic's parallel gated-frontier launch
• Gemini 3.5 Pro launch tracker — Google's June launch
• GPT-5 spec sheet — the model GPT-5.6 will replace