Skip to content

Models & licenses

Every model bundled or served by feder8d, together with its license and any attribution requirement.

Auto-generated from deploy/config/models.yaml.

Model pools

Plans grant access to pools, not individual models. The recommender chooses the best member.

small_pool

Small chat + small embedding tier. Available on Starter+.

  • qwen2.5-7b-instruct
  • qwen2.5-1.5b-instruct
  • llama3.2:3b
  • llama3.1:8b
  • moondream
  • bge-small-en-v1.5
  • bge-reranker-v2-m3
  • bge-m3
  • qwen2.5-coder:7b
medium_pool

Medium chat tier. Available on Pro+.

  • mistral-small-24b-instruct-2501
  • bge-large-en-v1.5
large_pool

Large frontier-class chat. NOT part of the managed inference plane at launch — every member is BYO-only (customer brings their own provider key, per-collection opt-in, prompt leaves the feder8d boundary with a console warning). Self-hosted large inference in eu-central-1 is a Phase-2 trigger. Entitled on Business+ as BYO; included managed on Dedicated.

  • llama-3.3-70b-instruct
reasoning_pool

Reasoning-specialised models — DeepSeek-R1 distillations. Auto-routed when the recommender sees `signals.workload in {code, math, plan}` and the tenant carries `entitlements.reasoning_models_enabled=true`. Gated Business+ + Dedicated.

  • deepseek-r1-distill-qwen-32b
  • deepseek-r1-distill-qwen-7b
image_pool

Text-to-image generation. Served by a diffusion backend (ComfyUI / Diffusers) behind the gateway's /v1/images/generations route, NOT vLLM / Ollama. Self-hosted = sovereign: prompts + generated images never leave the tenant boundary. Entitled on Business+ / Dedicated.

  • flux-schnell

Chat models

Model Params License Attribution Provider Languages
Qwen2.5-1.5B-Instruct 1.5B Apache-2.0 self_hosted en
Llama-3.2-3B-Instruct (Ollama) 3B Llama-3.2-Community Built with Llama local
Llama-3.1-8B-Instruct (Ollama, local GPU) 8B Llama-3.1-Community Built with Llama local
Qwen2.5-Coder-7B-Instruct (Ollama, local GPU) 7B Apache-2.0 local
Moondream 2 (vision, Ollama local GPU) 1.8B Apache-2.0 local
Qwen2.5-7B-Instruct 7B Apache-2.0 self_hosted en, fr, de, zh, es
Mistral-Small-24B-Instruct-2501 24B Apache-2.0 self_hosted en, fr, de, es, it, pt
Llama-3.3-70B-Instruct 70B Llama-3-Community Built with Llama external (together_ai) en, fr, de, es, it, pt, hi
Qwen2.5-32B-Instruct 32B Apache-2.0 self_hosted
DeepSeek-R1-Distill-Qwen-7B 7B Apache-2.0 self_hosted en, zh
DeepSeek-R1-Distill-Qwen-32B 32B Apache-2.0 self_hosted en, zh
DeepSeek-R1-Distill-Llama-70B 70B Llama-3-Community Built with Llama · Built with DeepSeek-R1 external (together_ai) en, fr, de, es, it, pt, hi
Kimi-K2 1T MoE (32B active) Modified-MIT Built with Kimi external (together_ai) en, zh, ja, ko, fr, de, es, it, pt
DeepSeek-V3 671B MoE (37B active) DeepSeek-License-v1.0 Powered by DeepSeek-V3 external (together_ai) en, zh, fr, de, es, it, pt, ja, ko
DeepSeek-R1 671B MoE (37B active) MIT Powered by DeepSeek-R1 external (together_ai) en, zh

Embedding models

ModelParamsLicenseProviderLanguages
BAAI/bge-m3 (Ollama, local) 568M MIT local
BAAI/bge-small-en-v1.5 33M MIT self_hosted en
BAAI/bge-large-en-v1.5 335M MIT self_hosted en

Reranker models

ModelParamsLicenseProviderLanguages
BAAI/bge-reranker-v2-m3 568M Apache-2.0 self_hosted en, af, fr, de, es, it, pt, zh, ja

Attribution notices

Where a model's license requires attribution, we display it in the response metadata and on this page.

Llama-3.2-3B-Instruct (Ollama) Built with Llama
Llama-3.1-8B-Instruct (Ollama, local GPU) Built with Llama
Llama-3.3-70B-Instruct Built with Llama
DeepSeek-R1-Distill-Llama-70B Built with Llama · Built with DeepSeek-R1
Kimi-K2 Built with Kimi
DeepSeek-V3 Powered by DeepSeek-V3
DeepSeek-R1 Powered by DeepSeek-R1

License text

The full text of each model's license is available from its source repository. We do not modify weights distributed under licenses that require source-of-modification disclosure.