Model Catalog

Instruction-tuned for 10 Indian languages. Strong for regional customer support and voice agent backends.

Flagship Indic chat model with extensive code-mixing support across 22 Indian languages.

High-throughput multilingual chat. Best economy option for English-first workloads with broad Indic coverage.

Google's latest multimodal model. Low-cost serving with native support for major Indian languages.

R1 chain-of-thought reasoning distilled from Qwen2.5-14B. Fits single A100 comfortably at FP16.

MoE design delivers frontier-quality reasoning at economy-tier throughput cost. 256K context window.

Strong coding and reasoning with hybrid thinking mode. High-context agentic and RAG workloads.

Best open reasoning model under 70B. Outperforms o1-mini on AIME 2024, MATH-500, and LiveCodeBench.

Highest-quality Indic model available. Dedicated GPU deployment — contact us to discuss fit.

Fast, accurate ASR across all 22 scheduled Indian languages including low-resource variants.

Whisper V3 fine-tuned on Vaani Hindi dataset. Handles diverse accents, noisy audio, and code-mixed speech.

State-of-the-art open ASR with unified offline and streaming inference. Language identification included.

Expressive TTS with 500+ speaker voices across 18 Indian languages. Prompt-controlled voice style.

44kHz TTS with zero-shot voice cloning and code-mixed text support across all 22 scheduled Indian languages.

MTEB-leading multilingual embeddings. Best for RAG over Indic and multilingual document corpora.

IBM's multilingual retrieval model with Matryoshka truncation and ONNX export. Minimal serving overhead.

Lightweight cross-encoder reranker for improving retrieval precision in multilingual RAG pipelines.

Encoder-only reranker with ~10x throughput advantage over decoder-based alternatives. Proven in production.