RWKV Project
8 models across 2 families · Latest: RWKV-7 Goose 2.9B (2025-03)
Linear complexity language models combining the efficiency of RNNs with the parallelism of Transformers
RWKV Project's portfolio covers 8 active models across 2 current families, spanning long context. Open a model detail page to compare provider routes and sourced benchmarks.
Covers 1 workload area across 8 active tracked models; last verified 2026-06-29.
Use it for
- Teams evaluating long context across this lab's releases
- Comparing model families before committing to a flagship
- Migration and pricing follow-ups across 8 tracked models
Do not use it for
- Choosing a hosting provider without opening a model page for price ladders
Active models
8
Current models from this lab, excluding deprecated ones
Active families
2
Current model families from this lab
Open catalog
8 open
8 open source / 0 open weights
Lowest output price
Not tracked
No provider output pricing linked yet
Latest dated release
2025-03-18
RWKV-7 Goose 2.9B
Freshness
2026-06-29
Researched 100d ago
Information
Release cadence
- Dated releases
- 5
- Latest model
- RWKV-7 Goose 2.9B
- Latest date
- 2025-03-18
Where this lab wins
- Long-context: 8 tracked models with context-token or InfiniteBench-class signal.
Flagship quality / price signal
- Flagship
- RWKV-6 Finch 14B
- Selection source
- Best sourced coding Q/$
- Coding grade
- unknown
- Benchmark / output price
- Not enough sourced benchmark and price coverage
About
The RWKV Project, maintained under the Linux Foundation AI & Data Foundation and led by Bo Peng, develops the RWKV (Receptance Weighted Key Value) family of language models. RWKV is a pure recurrent architecture that achieves linear O(n) time complexity during training and O(1) constant-memory inference — unlike Transformers which require quadratic attention and growing KV caches. The architecture has progressed through major versions: RWKV-4 (Dove, 2023), RWKV-5 (Eagle), RWKV-6 (Finch, 2024), RWKV-7 (Goose, 2025), and experimental RWKV-8 (Heron). All production models are released under Apache 2.0. The World series models are trained on multilingual corpora covering 100+ languages.
Featured models
| Model | Released | Context | Input price ($/1M) | Output price ($/1M) | License | Openness |
|---|---|---|---|---|---|---|
| RWKV-7 Goose 2.9B | 2025-03-18 | Infinite | - | - | Apache 2.0 | Open source |
| RWKV-7 Goose 1.5B | 2025-03-18 | Infinite | - | - | Apache 2.0 | Open source |
| RWKV-7 Goose 0.4B | 2025-03-18 | Infinite | - | - | Apache 2.0 | Open source |
Model families
Recent releases
- RWKV-7 Goose 2.9B- 2025-03-18
- RWKV-7 Goose 1.5B- 2025-03-18
- RWKV-7 Goose 0.4B- 2025-03-18
- RWKV-7 Goose 0.1B- 2025-03-18
- RWKV-6 Finch 14B- 2024-09-03
Explore related pages
Last reviewed: 2026-06-29. Data sourced from public lab announcements and provider documentation.