NEAR AI is the private AI and agent infrastructure layer around the NEAR ecosystem: model access, confidential inference, verifiable privacy, and agent execution surfaces such as IronClaw. NearStats tracks the public NEAR AI model catalog and direct confidential inference endpoints, then separates TEE/attested models from third-party gateway models.
AI models
47
Confidential / attested
21
Direct endpoints
18
Direct endpoint models
19
Reasoning models
24
Tool-capable models
41
Max context
1.05M tokens
Ready confidential
9
Live source check: NEAR AI model catalog is ok; direct endpoint catalog is ok.
GLM-5.2 on NEAR AI
| Signal | Value |
|---|---|
| Status | Live |
| Model | GLM 5.2 (z-ai/glm-5.2) |
| Privacy tier | TEE confidential |
| Context | 500K tokens |
| Max output | 131K tokens |
| Price | $1.40/1M input / $4.40/1M output |
| Endpoint | glm-5-2.completions.near.ai |
| Features | json_mode, reasoning, structured_outputs, tools |
Confidential model signals
| Metric | Value |
|---|---|
| TEE / attested catalog models | 21 |
| Third-party gateway models | 26 |
| Direct endpoint domains | 18 |
| Ready confidential models | 9 |
| Lowest-priced ready confidential text model | Gemma 4 31B Instruct at $0.13/1M input |
| Largest context model | GPT-5.4 with 1.05M tokens |
Featured NEAR AI models
| Model | Privacy tier | Context | Input / Output | Endpoint |
|---|---|---|---|---|
| GLM 5.2 (nearai) | TEE confidential | 500K tokens | $1.40/1M / $4.40/1M | glm-5-2.completions.near.ai |
| GPT-5.4 (openai) | Third-party gateway | 1.05M tokens | $2.50/1M / $15.00/1M | gateway |
| Gemma 4 31B Instruct (nearai) | TEE confidential | 262K tokens | $0.13/1M / $0.40/1M | gemma-4-31b.completions.near.ai |
| DeepSeek V4 Flash (nearai) | TEE confidential | 1.05M tokens | $0.17/1M / $0.35/1M | dsv4-flash.completions.near.ai |
| Qwen 3.6 27B FP8 (nearai) | TEE confidential | 262K tokens | $0.325/1M / $3.25/1M | qwen3-6-27b.completions.near.ai |
| Qwen 3.6 35B A3B FP8 (nearai) | TEE confidential | 262K tokens | $0.17/1M / $1.10/1M | qwen3-6-35b.completions.near.ai |
| Qwen3.5 122B A10B (nearai) | TEE confidential | 262K tokens | $0.40/1M / $3.20/1M | qwen35-122b.completions.near.ai |
| GLM 5.1 (nearai) | TEE confidential | 203K tokens | $0.85/1M / $3.30/1M | glm-5-1.completions.near.ai |
| GPT OSS 120B (nearai) | TEE confidential | 131K tokens | $0.15/1M / $0.55/1M | gpt-oss-120b.completions.near.ai |
| Qwen3-VL-30B-A3B-Instruct (nearai) | TEE confidential | 16K tokens | $0.15/1M / $0.55/1M | qwen3-vl-30b.completions.near.ai |
| FLUX.2-klein-4B (nearai) | TEE confidential | 128K tokens | $1.00/1M / $1.00/1M | flux2-klein.completions.near.ai |
| Qwen3-Embedding-0.6B (nearai) | TEE confidential | 41K tokens | $0.01/1M / $0.01/1M | qwen3-embedding.completions.near.ai |
Direct confidential endpoints
| Direct endpoint | Models |
|---|---|
| dsv4-flash.completions.near.ai | deepseek-ai/DeepSeek-V4-Flash |
| flux2-klein.completions.near.ai | black-forest-labs/FLUX.2-klein-4B |
| gemma-4-31b-int4-dsv4-test.completions.near.ai | google/gemma-4-31B-it-INT4-AutoRound |
| gemma-4-31b.completions.near.ai | google/gemma-4-31B-it |
| glm-5-1.completions.near.ai | zai-org/GLM-5.1-FP8 |
| glm-5-2.completions.near.ai | z-ai/glm-5.2, zai-org/GLM-5.2-FP8 |
| glm-5.completions.near.ai | zai-org/GLM-5-FP8 |
| gpt-oss-120b.completions.near.ai | openai/gpt-oss-120b |
| privacy-filter.completions.near.ai | openai/privacy-filter |
| qwen3-30b.completions.near.ai | Qwen/Qwen3-30B-A3B-Instruct-2507 |
| qwen3-6-27b.completions.near.ai | Qwen/Qwen3.6-27B-FP8 |
| qwen3-6-35b-nvfp4.completions.near.ai | Qwen/Qwen3.6-35B-A3B |
| qwen3-6-35b.completions.near.ai | Qwen/Qwen3.6-35B-A3B-FP8 |
| qwen3-embedding.completions.near.ai | Qwen/Qwen3-Embedding-0.6B |
| qwen3-reranker.completions.near.ai | Qwen/Qwen3-Reranker-0.6B |
| qwen3-vl-30b.completions.near.ai | Qwen/Qwen3-VL-30B-A3B-Instruct |
| qwen35-122b.completions.near.ai | Qwen/Qwen3.5-122B-A10B |
| whisper-large-v3.completions.near.ai | openai/whisper-large-v3 |
Provider mix
| Provider / owner | Models |
|---|---|
| openai | 15 |
| nearai | 14 |
| attested 3p | 7 |
| anthropic | 5 |
| 5 | |
| qwen | 1 |
How NEAR AI private inference works
NEAR AI private inference uses Trusted Execution Environments so prompts and responses can be processed inside isolated hardware-backed environments. NearStats links the public docs rather than re-verifying individual requests from the dashboard: see Private Inference, Verification, and the NEAR AI Cloud model catalog.
Privacy tiers NearStats shows
| Tier | Meaning |
|---|---|
| TEE confidential | A NEAR AI / attested model also exposed through a direct completions.near.ai endpoint. |
| Attested | A NEAR AI or attested third-party model in the public catalog, without a matched direct endpoint domain in the current endpoint list. |
| Third-party gateway | A third-party model surfaced through NEAR AI, tracked separately because it is not the same privacy tier as direct confidential inference. |
Frequently asked questions
- Does NEAR run AI models on-chain?
- NearStats tracks NEAR AI Cloud model infrastructure, private inference, direct completions endpoints, and agent infrastructure around the NEAR ecosystem. It does not imply that large models execute directly on NEAR blockchain validators.
- What counts as a confidential NEAR AI model on NearStats?
- NearStats counts NEAR AI catalog models owned by nearai or attested 3p as confidential or attested, and separately counts direct completions.near.ai endpoint domains.
- Are OpenAI, Anthropic, and Google models confidential in the same way?
- No. Third-party gateway models are separated from NEAR AI TEE-hosted or attested models, because upstream providers have different privacy guarantees than direct confidential inference endpoints.
Machine-readable data: /api/near/ai. Source policy: NearStats does not use Dune or social posts for this page's live numbers; it uses public NEAR AI endpoints and caches the normalized response server-side.
LeoDex routeSwap with NEAR Intents
Swap