OpenAI
OpenAI has 33 model versions on record; 18 have at least one documented fact.
Of the 18 versions listed below, 18 document a published framework; 12 document threshold determinations; 7 document tests by an independent or government party; 14 document tests the provider ran itself; 3 document safeguards the provider describes.
Model versions with documented facts
| Model | What is documented |
|---|---|
| GPT-6.1 Sol · Released 29 September 2026 · 1 document | |
| GPT-Image-2.5-Flare · Released 8 September 2026 · 1 document | |
| GPT-6 Astra · Released 3 September 2026 · 12 documents | |
| GPT-6 Sol · Released 3 September 2026 · 1 document | |
| GPT-5.6 Luna · Released 9 July 2026 · 1 document | |
| GPT-5.6 Sol · Released 9 July 2026 · 18 documents | |
| GPT-5.6 Terra · Released 9 July 2026 · 2 documents | |
| GPT-Rosalind-5.5 · Released 3 June 2026 · 1 document | |
| GPT-5.5 Instant · System card dated 4 May 2026 · 1 document | |
| GPT-5.5 · Released 23 April 2026 · 19 documents | |
| GPT-5.5 Pro · Released 23 April 2026 · 2 documents | |
| GPT-5.4 Thinking · Released 5 March 2026 · 1 document | |
| GPT-5.4 mini · Released 5 March 2026 · 6 documents | |
| GPT-5.3-Codex · Released 5 February 2026 · 2 documents | |
| GPT-5 · System card dated 13 August 2025 · 14 documents | |
| o3 · Released 16 April 2025 · 7 documents | |
| o4-mini · Released 16 April 2025 · 2 documents | |
| Operator · Released 23 January 2025 · 1 document |
No document we read states a capability level, a test or a safeguard for: ChatGPT Images 2.5, GPT-5.6 Sol (Preview), GPT-6 Luna (October), GPT-6 Sol (October), GPT-Live-1.1, GPT-Live-1.1 Quip. GPT-Image-2.5-Sunburst, GPT-6 Luna, GPT-Live-1, GPT-Live-1 mini, ChatGPT Images 2.0, ChatGPT Images 2.0 Thinking, GPT-5.3 Instant, GPT-5.6 Luna (August), GPT-5.6 Sol (August): their documents report tests only as table rows or lines that do not name the version on their own; each page links the documents.
Framework
2 of 2 framework documents listed here are stored as "proposed": a model read their level names and no person has confirmed them.
- Preparedness Framework, v2.0
Levels named in the stored text
- Critical (Cyber)
- High (Autonomy & replication)
- Frontier Governance Framework
Levels named in the stored text
- TIER 1 (Chemical, biological, radiological and nuclear)
- TIER 2 (Chemical, biological, radiological and nuclear)
- TIER 3 (Chemical, biological, radiological and nuclear)
Each test and safeguard, by who documents it
For each item, how many of the 18 versions listed above whose documents we read in full have it from an outside party (an independent or government tester), from the provider, or as a claim we could not confirm.
- Published framework with capability thresholds (Governance): provider 18
- Threshold determination for this release (Governance): provider 13
- Pre-release external access (Governance): outside party 1
- Tested: chemical, biological, radiological, nuclear (Testing): outside party 6; provider 4
- Tested: cyber capabilities (Testing): outside party 3; provider 6
- Tested: autonomy and replication (Testing): outside party 2; provider 2
- Tested: deception and sabotage (Testing): outside party 3; provider 3
- Tested: misuse and jailbreak (Testing): outside party 3; provider 2
- Tested: harmful content (Testing): outside party 1; provider 1
- Tested: injection and agentic risks (Testing): outside party 1; provider 4
- Tested: bias (Testing): provider 2
- Tested: privacy (Testing): provider 1
- Safety training (harmlessness and refusal) (Safeguards): provider 3
No document we read states, for any of these versions: Serious-incident disclosure commitment; Agent permission limits and sandboxing; Human approval for high-risk actions; Misuse monitoring and enforcement; Reasoning (chain-of-thought) monitoring; Staged or restricted access; Output provenance and watermarking; Telemetry and audit logging; Domain-specific safeguards (CBRN or cyber); Vulnerability disclosure and bug-bounty programme; Third-party or government red-team access during development.