MORENA 1.5B Instruct
Vambo AI's African foundation model for chat, translation, and tool use across 12 African languages.
African languages · Apache 2.0Open-model inference · Private beta
Open-model inference, built for teams serving Africa.
Call Qwen, Mistral, Gemma and African open models through one OpenAI-compatible API. Start on Yamify-managed capacity and choose verified African regions as they become available.
Applications are reviewed by the founding team. No infrastructure is provisioned until access is approved and funded.
One APISwitch open models without rewriting your application
African data planeModel serving designed to run physically inside Africa
Verifiable regionEndpoint region now; signed request receipts on the roadmap
Trending open-weight models
The models teams are asking for. One API.
Start with efficient shared inference. Move to a dedicated African deployment when traffic, latency, or data residency requires it.
Ministral 3 8B
Mistral's compact open-weight model for multilingual assistants and application workflows.
Multilingual · Apache 2.0 · Runtime testing pendingGemma 4 12B
Google's efficient open-weight model for multilingual assistants, document workflows, and production apps.
Efficient · MultilingualQwen 3.5
Fast general-purpose reasoning, code, agents, and tool calling through Yamify's shared model pool.
Reasoning · Code · AgentsWhisper Large v3
Transcribe multilingual audio into text or translate speech into English.
OpenAI · Speech to text · Apache 2.0VibeVoice Realtime 0.5B
Generate single-speaker speech from streaming text. English-first; other languages are experimental.
Microsoft · Text to speech · MITAvailability and serving region are shown before you send production traffic.
Audio models are catalog previews. Speech API access and pricing are not available yet.
Explore 34 more self-hostable model candidates · Coming soon
Selected from Together AI and DeepInfra after checking publisher weights, license metadata and conservative A100 40GB sizing. These are planned models—not working Yamify endpoints. Runtime testing and prices are still pending.
Each entry below is a single-A100 candidate. Up to eight servers can provide separate capacity; their VRAM is not automatically pooled. The whole catalog will not run simultaneously.
34 models · Pricing after benchmarking
- COMING SOON · Chat & vision
granite-4.2-8b
ibm-granite · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 16.4 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Chat & vision
Mistral-Nemo-Instruct-2407
mistralai · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 22.9 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Chat & vision
phi-4
microsoft · mit · 1 × A100 40GB candidate
Estimated weights: 27.4 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Chat & vision
Qwen3-14B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 27.6 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Chat & vision
Qwen3.5-9B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 18 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Classification
clef-flash
Cloudflare · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 17.6 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
all-MiniLM-L12-v2
sentence-transformers · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 0.1 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
all-MiniLM-L6-v2
sentence-transformers · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 0.1 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
all-mpnet-base-v2
sentence-transformers · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 0.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
bge-base-en-v1.5
BAAI · mit · 1 × A100 40GB candidate
Estimated weights: 0.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
bge-en-icl
BAAI · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 13.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
bge-large-en-v1.5
BAAI · mit · 1 × A100 40GB candidate
Estimated weights: 0.7 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
clip-ViT-B-32-multilingual-v1
sentence-transformers · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 0.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
e5-base-v2
intfloat · mit · 1 × A100 40GB candidate
Estimated weights: 0.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
e5-large-v2
intfloat · mit · 1 × A100 40GB candidate
Estimated weights: 0.7 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
embeddinggemma-2
google · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 1.4 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
gte-base
thenlper · mit · 1 × A100 40GB candidate
Estimated weights: 0.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
gte-large
thenlper · mit · 1 × A100 40GB candidate
Estimated weights: 0.7 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
multilingual-e5-large
intfloat · mit · 1 × A100 40GB candidate
Estimated weights: 1.1 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
multilingual-e5-large-instruct
intfloat · mit · 1 × A100 40GB candidate
Estimated weights: 1.1 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
paraphrase-MiniLM-L6-v2
sentence-transformers · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 0.1 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
Qwen3-Embedding-0.6B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 1.2 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
Qwen3-Embedding-4B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 7.5 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
Qwen3-Embedding-8B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 14.1 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Embeddings
text2vec-base-chinese
shibing624 · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 0.2 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Reranking
Qwen3-Reranker-0.6B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 1.2 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Reranking
Qwen3-Reranker-4B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 7.5 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Reranking
Qwen3-Reranker-8B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 15.3 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Speech to text
Qwen3-ASR-0.6B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 1.8 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Speech to text
Qwen3-ASR-1.7B
Qwen · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 4.4 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Speech to text
Voxtral-Mini-3B-2507
mistralai · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 8.8 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Speech to text
whisper-large-v3
openai · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 2.9 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Speech to text
whisper-large-v3-turbo
openai · mit · 1 × A100 40GB candidate
Estimated weights: 1.6 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗ - COMING SOON · Text to speech
Audio8-TTS-Preview-0.6b
Audio8 · apache-2.0 · 1 × A100 40GB candidate
Estimated weights: 1.2 GiB at BF16/FP16, plus runtime memory. Context and batch limits will be tested before launch.
Publisher model card ↗