AIREITERGPT-6 has no model card, API endpoint, or release date as of August 2026. The current flagship is GPT-5.6, shipped July 9, 2026 (Sol/Terra/Luna tiers). But the release cadence has accelerated - GPT-5.5 followed GPT-5 by eight months, and GPT-5.6 followed just eleven weeks later. That acceleration, combined with a Sam Altman comment and prediction-market activity, makes GPT-6 a question of "when" more than "if."
What happened to "Spud" - and why it reshapes the timeline
Throughout early 2026, the codename "Spud" circulated as the rumored identity of GPT-6. ZeroTwo reported that Spud completed pretraining on March 24, 2026, at the Stargate data center in Abilene, Texas, with an estimated 2x+ the training compute of GPT-5 per Epoch AI analysis.
But Spud didn't become GPT-6. Voiceflow reported that it shipped as GPT-5.5 on April 23, 2026 - the model many expected to be "6" was a versioned increment, not a generational jump.
The release cadence tells the story:
| Model | Release date | Gap from previous |
|---|---|---|
| GPT-5 | August 7, 2025 | - |
| GPT-5.5 | April 23, 2026 | ~8 months |
| GPT-5.6 | July 9, 2026 | ~11 weeks |
| GPT-6 | Not announced | - |
The intervals are compressing. Voiceflow reported that GPT-5.5 introduced persistent memory, personalization, and 1M-token context. GPT-5.6 then added multi-agent ultra mode (four parallel agents by default), programmatic tool calling, and a 52.7% score on Agents' Last Exam.
On April 28, 2026, four days after GPT-5.5 launched, Sam Altman joked that GPT-6 should ship "with extra goblins" during a discussion about a Codex personality bug. He gave no timeline, specs, or roadmap.
What GPT-6 is expected to bring
No GPT-6 feature is confirmed by OpenAI. What follows is organized by confidence level, with sources for each claim.
Reported across multiple outlets (moderate confidence):
- Persistent cross-session memory. This is the highest-confidence expected feature. Voiceflow reports that Altman has said "people want memory" and "people want product features that require us to be able to understand them," and that GPT-5.5 introduced temporary memory he described as not yet encrypted. ZeroTwo calls memory "the headline feature" for the next generation.
- Native agentic execution. Multi-step browsing, coding, file operations, and API calls without step-by-step prompting. ZeroTwo reports this as a core expected capability, estimating agentic tasks may consume 50K-500K tokens per job.
Rumored from single or unverified sources (low confidence):
- 2M-token context window. ZeroTwo reported this but explicitly marked it as low-confidence. For reference, Gemini 3.5 Pro already offers 2M tokens and GPT-5.6 Sol supports up to 1M.
- ~10T parameters. Kie.ai cited this as leak-only, with no verifiable source. GPT-5.6's parameter count is not publicly known.
Speculative (inference from current limitations, not sourced):
- Better behavioral consistency. The Codex personality bug that prompted the "extra goblins" joke illustrated how training reward signals can generalize into unexpected model behavior.
- Deeper personalization tied to persistent memory.
GPT-6 release signals worth tracking
Prediction markets offer one directional data point. Kie.ai reported Polymarket contracts from mid-July 2026:
| Deadline | Polymarket probability |
|---|---|
| July 31, 2026 | ~1% |
| August 14, 2026 | ~3% |
| August 31, 2026 | ~24% |
| September 30, 2026 | ~71% |
| December 31, 2026 | ~88% |
These are market signals, not forecasts, and "release" may mean anything from a limited preview to general API availability.
Real confirmation triggers to watch:
- An official OpenAI blog post or announcement page naming "GPT-6."
- An entry in the OpenAI API changelog with a gpt-6 model ID.
- A published model card or safety evaluation.
- Sam Altman or OpenAI leadership providing a concrete timeline in a public forum.
The bar GPT-6 needs to clear
GPT-6 won't launch into an empty field. The frontier has moved considerably since GPT-5's release twelve months ago. Benchmark and pricing data below are sourced from OpenAI's GPT-5.6 page, our Grok 4.5 analysis, and ZeroTwo's frontier comparison:
| Model | Key strengths | Context | API price (input/output per 1M tokens) |
|---|---|---|---|
| GPT-5.6 Sol | 52.7% Agents' Last Exam, 80 Coding Agent Index, 62.6% OSWorld | 1M tokens | $5 / $30 |
| Claude Opus 5 | Higher on GDPval-AA v2 (1,759.6 vs Sol's 1,747.8 Elo), strong reasoning | 1M tokens | $5 / $25 |
| Gemini 3.5 Pro | 2M context, strong multimodal | 2M tokens | Not publicly listed |
| Grok 4.5 | 83.3% Terminal Bench, fast output (~80 tok/s) | 500K tokens | $2 / $6 |
| Kimi K3 | Open weights, competitive coding | 256K tokens | Open-source (self-host) |
On the Artificial Analysis Coding Agent Index, GPT-5.6 Sol scores 80, above Claude Fable 5 at 77.2. For GPT-6 to be a meaningful generational step, it would need to deliver more than incremental benchmark gains - persistent memory and reliable multi-step autonomy are the two areas where user demand clearly outstrips current capability.
What to do while GPT-6 isn't here
Use GPT-5.6 today. GPT-5.6 Sol is available through the OpenAI API and through API providers that aggregate multiple models. Pricing starts at $5 input / $30 output per million tokens for Sol, with Terra ($2.50/$15) and Luna ($1/$6) as cheaper tiers. See our GPT-5.6 pricing guide for tier-by-tier cost analysis.
Build model-agnostic workflows and an evaluation harness. Orchestration frameworks like LangGraph, AutoGen, and CrewAI can reduce migration work when swapping models, though tool schemas, context limits, and structured-output behavior may still need adjustment. Pair that with a test pack of your hardest prompts and production tasks - when GPT-6 drops, you'll be able to benchmark it against GPT-5.6 Sol on day zero instead of guessing. Elser AI recommends tracking specific metrics: fewer retries to a usable draft, better instruction adherence, and less long-context drift.
FAQ
Is GPT-6 released yet?
No. As of August 2026, OpenAI has not announced, previewed, or released GPT-6. The most recent OpenAI model is GPT-5.6, released July 9, 2026.
When will GPT-6 come out?
No official date exists. Watch for an OpenAI blog post or an API changelog entry naming a gpt-6 model ID. Polymarket odds from mid-July 2026 suggest roughly 71% by September 30 and 88% by December 31, but these are market bets, not announcements.
Will GPT-6 be called GPT-6?
Unknown. OpenAI's recent naming pattern (GPT-5, GPT-5.5, GPT-5.6) suggests they may continue with incremental versioning rather than jumping to a new major number. The model widely expected to be "GPT-6" earlier in 2026 was released as GPT-5.5.
What will GPT-6 cost?
No pricing exists for GPT-6. For reference, GPT-5.6 Sol costs $5 input / $30 output per million tokens through the OpenAI API. ZeroTwo's speculative estimate put a potential GPT-6 price at $10-$30 input / $40-$80 output per million tokens, with a possible ChatGPT Pro tier at $200-$300/month - but these are guesses with no official backing.