Newest tt-model bundles
The 20 models with the most recently pushed tt-model bundles, newest first.
Community models are tt-model bundles published by their owners and are not in tt-inference-server's release spec. Tenstorrent has not reviewed them. Serve at your own risk.
clef
Clef is a 27B-parameter multimodal decision model from Cloudflare: a Qwen3.8-27B backbone (64 layers, 48 Gated DeltaNet and 16 full-attention, hidden size 5120) with its 27-block vision tower, and a 65M-parameter joint schema head.
- Runs on
- p150x2
- Status
- Not in the release spec
- tt launch
- None
clm-v0.1-8b
CLM-v0.1-8B is a System One verifier and ranker: a frozen Qwen3-8B encoder (36 layers, last-token pooling) with two 9.4M-parameter MLP projection heads (4096-1536-1536-512), a state head and an action head, trained with a bidirectional…
- Runs on
- BH 4xP150, p150
- Status
- Not in the release spec
- tt launch
- None
gemma-4-26B-A4B-it
Gemma 4 26B A4B instruction-tuned, a 25.2B-parameter mixture-of-experts model (3.8B active parameters; 30 layers; 128 experts with 8 active plus 1 shared), served with vLLM for chat, reasoning and tool calling on text input.
- Runs on
- BH 4xP150, p150, p150x2
- Speed
- 37 tok/s175 ms 1 chip
- Status
- Not in the release spec
- tt launch
- 7 clients
gpt-oss-120b
Listed in tt-inference-server on 4 hardware types; best status Experimental. 3 tt-model bundles also published.
- Runs on
- BH 4xP150, BH Galaxy, BH LoudBox +3
- Speed
- ≥ 2.7 tok/s≤ 510 ms BH QuietBox 2
- Status
- Experimental
- tt launch
- 7 clients
gpt-oss-20b
gpt-oss-20b, OpenAI's 21B-parameter mixture-of-experts reasoning model (3.6B active parameters), served with vLLM for chat, reasoning and function calling.
- Runs on
- BH 4xP150, BH LoudBox, p150 +1
- Speed
- 48 tok/s209 ms 1 chip
- Status
- Not in the release spec
- tt launch
- 7 clients
Laguna-XS-2.1
Laguna-XS-2.1, poolside's 33B-parameter mixture-of-experts coding model (3B active parameters), served with vLLM with tool calling and separated reasoning output.
- Runs on
- BH 4xP150, p150x2
- Speed
- 20 tok/s215 ms 2 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
laya
Laya is a 421M-parameter non-LLM calibrated decision model: a ModernBERT-large encoder (28 layers, 1024 hidden, GeGLU, dual-theta RoPE, one global attention layer per three) with a 2-layer decision head that has an option-marker scorer and…
- Runs on
- BH 4xP150, p150
- Status
- Not in the release spec
- tt launch
- None
laya-typed-decisions
Laya typed-decisions is the authors' fine-tune of Laya on the LocalLLaMA/typed-decisions train split: a 421M-parameter ModernBERT-large encoder (28 layers, 1024 hidden, GeGLU, dual-theta RoPE, one global attention layer per three) with a…
- Runs on
- BH 4xP150, p150
- Status
- Not in the release spec
- tt launch
- None
llm-jp-4.1-8b-thinking
llm-jp/llm-jp-4.1-8b-thinking (LLM-jp-4.1 8B, thinking) served by vLLM and the Tenstorrent vLLM plugin on one Blackhole P150 chip (p150a).
- Runs on
- p150
- Status
- Not in the release spec
- tt launch
- 7 clients
Muse-Glimmer-30B
Muse-Glimmer-30B, a dense transformer of about 29.6B parameters including its vision encoder, served with vLLM as a text-only endpoint for long-context agentic coding with tool calling and separated reasoning output.
- Runs on
- BH QuietBox 2
- Speed
- 42 tok/s70 ms 4 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
Olmo-3.1-32B-Instruct
Olmo 3.1 32B Instruct, a fully open 32B dense decoder with hybrid sliding-window / full attention and 65k context, served with vLLM.
- Runs on
- BH P300, BH QuietBox 2, p150
- Speed
- 11 tok/s177 ms 1 chip
- Status
- Not in the release spec
- tt launch
- 7 clients
Ornith-1.0-35B
Ornith-1.0-35B, a 35B-parameter mixture-of-experts model for agentic reasoning and coding, served with vLLM with tool calling and separated reasoning output.
- Runs on
- BH 4xP150, p100, p150 +1
- Speed
- 12 tok/s253 ms 1 chip
- Status
- Not in the release spec
- tt launch
- 7 clients
Ornith-1.5-9B
Ornith-1.5-9B, a 9B-parameter dense reasoning and coding model, served with vLLM with tool calling and separated reasoning output.
- Runs on
- BH 4xP150, p150, p150x2
- Speed
- 6.0 tok/s97 ms 1 chip
- Status
- Not in the release spec
- tt launch
- 7 clients
Qwen3.6-27B
Listed in tt-inference-server on 1 hardware type; best status Experimental. 1 tt-model bundle also published.
- Runs on
- BH 4xP150, BH QuietBox 2
- Status
- Experimental
- tt launch
- 7 clients
Qwen3.8-27B
Listed in tt-inference-server on 1 hardware type; best status Complete. 3 tt-model bundles also published.
- Runs on
- BH 4xP150, BH QuietBox 2, p150x2
- Speed
- ≥ 20 tok/s≤ 124 ms BH QuietBox 2
- Status
- Complete
- tt launch
- 7 clients
Qwen3.8-Flash-Next
Qwen3.8-Flash-Next, a mixture-of-experts model with 125B parameters (about 6B active per token) plus a 51B n-gram embedding table, served with vLLM as a text-only chat and completion endpoint with tool calling and separated reasoning…
- Runs on
- BH 4xP150, BH LoudBox, BH QuietBox 2
- Speed
- 24 tok/s340 ms 4 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
Ornith-1.5-122B
Ornith-1.5-122B, a 122B-parameter mixture-of-experts model with 10B parameters active per token (256 experts, top-8 plus a shared expert; 48 layers mixing Gated DeltaNet linear attention with gated full attention; the Qwen3.5-122B-A10B…
- Runs on
- BH 4xP150, BH QuietBox 2
- Speed
- 133 tok/s230 ms 4 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
Ornith-1.5-35B-A3B
Ornith-1.5-35B-A3B (35B total / 3B active MoE: 256 experts, top-8 + shared expert, 40 layers of Gated DeltaNet + gated attention, vision tower, MTP head) on Blackhole through vLLM and the Tenstorrent plugin, with a fused decode step: the…
- Runs on
- BH 4xP150, BH P300, BH QuietBox 2 +3
- Speed
- 220 tok/s118 ms 4 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
Qwen3-Coder-Next
Qwen3-Coder-Next (80B total / 3B active MoE: 512 experts, top-10 + shared expert, 48 layers of Gated DeltaNet + gated attention) on Blackhole through vLLM and the Tenstorrent plugin, with a fused decode step: the v1 bundle…
- Runs on
- BH 4xP150, BH P300, BH QuietBox 2 +1
- Speed
- 179 tok/s173 ms 4 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
Qwen3.5-122B-A10B
Qwen3.5-122B-A10B, a 122B-parameter mixture-of-experts model with 10B parameters active per token (256 experts, top-8 plus a shared expert; 48 layers mixing Gated DeltaNet linear attention with gated full attention) and a vision encoder,…
- Runs on
- BH 4xP150, BH QuietBox 2
- Speed
- 131 tok/s216 ms 4 chips
- Status
- Not in the release spec
- tt launch
- 7 clients
No model in this list matches the filter.