Research observatory 16 / Disclosure quality
AI Transparency Scorecard
Compares model developers by the launch fields they disclose, without substituting missing facts with estimates.
Public editionLive dataset
- Developers scored
- 6
11 verified model records
- Median disclosure
- 67%
6 tracked launch fields
- Fully disclosed
- 0
All tracked fields present
- Top disclosure
- 75%
Anthropic
Data status6 verified observations
Bar height represents field disclosure rate (%). Colors separate the recorded series.
Verified observationHover or focus any mark for exact valuesData and chart downloads share this dataset
Coverage note
Scores reflect fields captured in verified launch records and can change when better primary disclosures are added. They do not evaluate undisclosed internal practices.
- Dataset ID
- spd:transparency-scorecard
- Coverage
- Live collection
- Records
- 11
- Fields
- 10
- Formats
- CSV / JSON
- Updated
| Model | Developer | Fields disclosed | Disclosure score | license | availability | context Window Tokens | input Price Per Million | output Price Per Million | release Type |
|---|---|---|---|---|---|---|---|---|---|
| GPT-4b micro | OpenAI | 3 | 50 | — | developed for research purposes; not broadly available | 64000 | — | — | experimental research |
| GPT-5.2 | OpenAI | 3 | 50 | — | In ChatGPT: GPT-5.2 Instant, Thinking, and Pro begin rolling out today to paid plans; In the API: available now to all developers. | 262144 | — | — | API and ChatGPT (model variants: GPT-5.2 Instant, GPT-5.2 Thinking, GPT-5.2 Pro) |
| GPT Image 1.5 | OpenAI | 2 | 33 | — | Model rolling out in ChatGPT for all users and available in the API as GPT Image 1.5; the new Images experience is rolling out to most ChatGPT users, with Business and Enterprise access coming later; global. | — | — | — | API and ChatGPT rollout |
| Claude Mythos Preview | Anthropic | 5 | 83 | — | restricted access (restricted-access rollout) | 1000000 | 25.000000 | 125.000000 | Preview |
| Claude Fable 5 | Anthropic | 4 | 67 | — | Publicly accessible via API and Enterprise; included in subscription plans through 2026-06-22 after which access requires purchase of usage credits; some sensitive-topic access restricted and Mythos 5 limited to vetted Project Glasswing participants | — | 10.000000 | 50.000000 | Public release (Mythos-class); available via API and Enterprise |
| GPT-Live | OpenAI | 2 | 33 | — | Rolling out to ChatGPT users globally (ChatGPT Voice); API availability planned/coming soon. (Update 2026-07-31: SynthID watermarking and API verification access added for supported audio.) | — | — | — | voice model (full-duplex) |
| Kimi K3 | Moonshot | 4 | 67 | License requires companies operating a model-as-a-service business with > $20M total annual revenue (over 12 months) to enter a separate agreement with Moonshot before commercial use. | Weights published for anyone to download; model can be run locally on user hardware | 1000000 | — | — | open-weights release (weights published for public download) |
| Muse Glimmer | Meta | 4 | 67 | Apache 2.0 | Consumer hardware with 24 GB or 32 GB of memory | 131072 | — | — | open weights |
| Nemotron 3.5 Lightning | Nvidia | 4 | 67 | OpenMDW-1.1 | Available through Nvidia and model repositories including Hugging Face and ModelScope; NeMo Switchyard available on GitHub | 1000000 | — | — | open weights (public release) |
| Qwen3.8-27B-CRACK-GGUF | Jinho Jang (dealign.ai) | 3 | 50 | — | Hugging Face repository; downloadable GGUF files and a 0.9 GB vision projector for local use | 262000 | — | — | Open weights; GGUF quantizations packaged for local inference via llama.cpp |
| GPT-5.6 Sol | OpenAI | 2 | 33 | — | Limited release to a small group of customers; wider access planned later | — | — | — | API |
Methodology
How to read this report
- 01The score measures whether six practical launch fields are disclosed: license, availability, context window, input price, output price, and release type.
- 02Every field receives equal weight; missing is distinct from zero and receives no credit.
- 03This is a disclosure-completeness score, not a safety, quality, or trustworthiness rating.
Sources
Evidence
9 publishers supporting 12 records. Expand a publisher to inspect its cited pages.