THE AI LANDSCAPE 03 / ACCESS & CONTROL

How open
is open AI?

Look beyond the label. Compare downloadable weights, model licenses, and deployment choices across selected releases.

THE BIG PICTURE

AI Openness

Selected coverage. Check each entry for context and evidence.

8 entries

Each entry links to its primary source ↗

AI Openness. Reviewed 2026-09-23. Editorial classifications, not a performance ranking.
Model / developerWeightsTerms shownDeploymentWhat to check
DeepSeek V4.1 FlashDeepSeek1 source DownloadableOfficial checkpointMITModel card and repositorySelf-host + APIChoose infrastructureCompute requirementsLarge checkpoint
Qwen3.8-27BQwen / Alibaba1 source Downloadable27B checkpointApache 2.0Official model cardSelf-hostOr a compatible providerExact model variantDo not generalize to all Qwen
Gemma 4Google DeepMind1 source DownloadableFour release sizesApache 2.0Gemma 4 release termsDevice or serverDepends on model sizeGeneration-specific termsEarlier Gemma terms differ
gpt-oss-20b / 120bOpenAI1 source DownloadableOpen-weight checkpointsApache 2.0Also review usage policySelf-host / third partySeparate from ChatGPTRelease: Aug 2025Not the hosted GPT-6 family
Qwen3.8 Flash Next FP8Qwen / Alibaba1 source DownloadableOfficial FP8 checkpointQwen Community 1.0Developer-specific termsSelf-host, subject to termsReview commercial scopeService use conditionsRead the full license
Llama 4 ScoutMeta2 sources DownloadableOfficial weightsLlama 4 CommunityDeveloper-specific termsSelf-host, subject to termsLicense and use policy applyEligibility and restrictionsReview the linked agreement
GPT-6 Sol / LunaOpenAI1 source Not offered in releaseService-based accessService termsNo weight license offeredAPI · Work · CodexProvider-hostedProduct accessModel access varies by surface
Claude Opus 5.5Anthropic1 source Not offered in releaseService-based accessService termsNo weight license offeredAPI · Claude · cloudsProvider-hostedCloud and product termsHosting choice is not weights
Downloadable / permissive licenseCustom license conditionsHosted serviceRead the classification notes ↗

A manually reviewed snapshot as of September 23, 2026. Colors organize the evidence; they do not score quality. On smaller screens, swipe across the comparison.

CONTEXT MAKES THE DIFFERENCE

How to read this landscape.

Our interpretation of the comparison.

01

Weights are one part of openness

A downloadable checkpoint does not tell you whether training data, training code, and the full development process are available.

02

Check the exact checkpoint

A developer can release both hosted models and downloadable models, or use different licenses within one family.

03

Control comes with operations

Self-hosting gives you deployment choices. It also means arranging hardware, serving software, monitoring, and updates.

TRANSPARENT BY DESIGN

Follow the evidence.

Official announcements, product documentation, and model repositories. All linked sources were checked on September 23, 2026. Publication dates and review dates are shown separately.

9primary sources
behind this comparison
Suggest a correction
Scope & methodology

This page compares distribution and the licenses linked by the named official repositories. “Permissive” groups checkpoints labeled Apache 2.0 or MIT; “Custom” groups developer-specific weight licenses; “Hosted only” means no downloadable weights are offered in the cited release. These categories are not a certification of open-source AI. License summaries are deliberately limited: read the linked terms for your exact version and use case. Older available checkpoints remain useful comparisons and are not presented as new releases.

Dates on continuously updated documentation are labeled “Current documentation” unless a specific publication date is available. This page is maintained manually; the review date is not a live-feed timestamp. Company announcements describe the company’s claims and plans, not an independent audit.

DeepSeek V4.1 Flash1 source

DeepSeek / Permissive license

DeepSeek V4.1 Flash

The official model card labels the release MIT. Its API is one access route; downloading the weights is a separate deployment choice.

Qwen3.8-27B1 source

Qwen / Alibaba / Permissive license

Qwen3.8-27B

The 27B checkpoint uses Apache 2.0. This classification belongs to this checkpoint, not to every model carrying the Qwen name.

Gemma 41 source

Google DeepMind / Permissive license

Gemma 4

Google released Gemma 4 in E2B, E4B, 26B MoE, and 31B Dense sizes under Apache 2.0. The April 2026 release remains a useful on-device and server comparison.

gpt-oss-20b / 120b1 source

OpenAI / Permissive license

gpt-oss-20b / 120b

OpenAI released these reasoning checkpoints in August 2025. They are distinct from the hosted GPT-6 products and can be deployed on infrastructure chosen by the user.

Qwen3.8 Flash Next FP81 source

Qwen / Alibaba / Custom license

Qwen3.8 Flash Next FP8

The official FP8 model card links Qwen Community 1.0, rather than Apache 2.0. Review its commercial service conditions before treating it like the 27B checkpoint.

Llama 4 Scout2 sources

Meta / Custom license

Llama 4 Scout

Scout uses Meta’s Llama 4 Community License. Weight availability does not remove the agreement’s requirements or make this an Apache-licensed release.

GPT-6 Sol / Luna1 source

OpenAI / Hosted only

GPT-6 Sol / Luna

The Sol and Luna announcement distributes these models through OpenAI services. It does not offer model weights for self-hosting.

Claude Opus 5.51 source

Anthropic / Hosted only

Claude Opus 5.5

Access through multiple cloud platforms provides more service choices. It does not make Opus 5.5 a downloadable open-weight model.

PRIMARY SOURCES

Reviewed 2026-09-23 · AIstify Research