
Gnani AI models
Gnani AI builds voice AI for Indian enterprises on its own models, including the Prisma speech-to-text, Timbre text-to-speech and Evon language models.
Bengaluru, IndiaFounded 2016gnani.ai
- Releases
- 6
- Open weights
- 1
- Latest
- Oct 28, 2026
October 2026
2 releasesGnani Prisma v3.0
Prisma v3.0 is Gnani's speech recognition model for nine Indian languages and English, built for code-mixed phone calls and offering four selectable output modes.
- Modalities
- audio → text
- At launch
- API, Enterprise
- Gnani reported a macro WER of 16.93 on Vistaar, against 19.98 for ElevenLabs Scribe v2 and 20.51 for Microsoft Azure STT
- Verbatim, Transcribe, Code-mix and Transliterate modes, chosen per request
- Keyterm prompting with up to 200 terms, plus language hints or automatic language detection
- Accepts 8 kHz telephony and 16 kHz audio and could run in the cloud, in hybrid setups or fully on premises
Gnani Timbre v3.0
Timbre v3.0 is Gnani's expressive text-to-speech model for Indian languages, adding inline audio tags that direct delivery such as laughs, sighs, whispers and pauses.
- Modalities
- text → audio
- At launch
- API
- Launched with 13 voices, 13 audio tags and 10 languages
- Reads amounts and dates the way a person would say them
- Available in the Gnani app at launch, with Pipecat and LiveKit support announced for later
- Timbre 2.5 remained in service for everyday voice agent conversations
August 2026
1 releaseGnani Evon v3.3
Evon v3.3 is Gnani's open-weights hybrid Mamba-Transformer mixture-of-experts reasoning model for English and Indian languages, adapted from NVIDIA's Nemotron 3 Nano with a rebuilt Indic tokenizer.
- Modalities
- text → text
- Context
- 128K context
- Parameters
- MoE: 30B total, ~3.5B active
- License
- Apache-2.0
- At launch
- Open weights
- About 3.5B parameters active per token and a 128K-token context window, sized to run on a single node
- Extended 163,072-token vocabulary that Gnani said cut tokens per Indian-language word by 38.6%
- Gnani reported a MILU macro score of 78.74 across 11 languages, ahead of Sarvam-105B at 75.71
- Launched as part of the Gnani Artha sovereign AI stack, with weights available by request on Hugging Face
June 2026
1 releaseGnani Prisma v2.5
Prisma v2.5 is Gnani's speech-to-text model for Indian languages, trained on 14 million hours of proprietary Indic speech and tuned for noisy, code-mixed telephony audio.
- Modalities
- audio → text
- At launch
- API
- Gnani reported first place in 8 of 9 Indian languages on real-world and noisy benchmarks, and 27 of 48 evaluations overall
- Gnani reported 15% lower WER for rural Hindi and 18% lower WER across Dravidian languages
- Handles GSM and VoIP codecs and word-level code-switching without language tags
- Gnani said post-training doubled throughput over the previous version
December 2025
1 releaseVachana STT
Vachana STT is Gnani's Indic speech-to-text model, trained on more than one million hours of real-world voice data and released under the IndiaAI Mission as the first piece of its VoiceOS roadmap.
- Modalities
- audio → text
- At launch
- API
- Gnani reported 30 to 40 percent lower word error rates on low-resource Indic languages and 10 to 20 percent lower on the top eight
- Handles telephony audio from 8 kbps to 64 kbps, with real-time and batch transcription
- Gnani said it was processing about 10 million calls a day with p95 latency of 200 ms
- Offered through enterprise API access, with 100,000 free minutes for early adopters at launch
October 2024
1 releaseGnani speech-to-speech LLM
Gnani's speech-to-speech LLM is a voice model for automated customer conversations, built on NVIDIA's AI platform and trained on more than 14 million hours of multilingual conversational data.
- Modalities
- audio → audio
- At launch
- Enterprise
- Supported 14 languages at launch, per Gnani
- Gnani said it handled more than 10 million voice interactions a day at launch
- Built with NVIDIA NeMo and served with TensorRT-LLM, Triton and Riva
- Aimed at enterprise voice automation in sectors such as finance and healthcare
Cite and reuse this data
These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL.
Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".
Citation
The AI Engineering Editors. "Gnani AI models: release history and timeline." The AI Engineering, updated Oct 28, 2026. https://theaiengineering.com/releases/gnani-ai
HTML link
<a href="https://theaiengineering.com/releases/gnani-ai">Gnani AI models: release history and timeline</a> (The AI Engineering)- Full dataset
- 1,250 releases from 73 labs
- Coverage
- to
- Updated
- License
- CC BY 4.0
- Publisher
- The AI Engineering
Other labs
Google106
Alibaba Qwen84
OpenAI79Mistral AI59
Meta44
- Z.ai43
Microsoft42Tencent42
- ByteDance Seed40
- DeepSeek38
- Liquid AI34
Cohere31- IBM30
xAI30- NVIDIA29
MiniMax28
Stability AI28- Anthropic26
Ai223StepFun22
Amazon21- Baidu21
- ElevenLabs21
- Sarvam AI21
Black Forest Labs18
- Kuaishou16
- Meituan16
Runway16- Moonshot AI15
Midjourney14
BharatGen13Salesforce13
- Xiaomi13
Nous Research11- Perplexity10
AI4Bharat9
Krutrim9
AI21 Labs8- Apple8
Ideogram8
01.AI7
Hugging Face7
Lightricks7Luma AI7
Reka7- Suno7
- Databricks5
Leonardo.Ai5- Pika5
Recraft5- Mannas4
Soket AI4
Fractal3
Inflection AI3MBZUAI IFM3
- Snowflake3
Thinking Machines Lab3- Infosys2
NPCI2
Tech Mahindra2
TypeSafe AI2
ZenteiQ2ARTPARK (IISc)1
Avataar1- Cognition1
CoRover.ai1
Genmo1Josh Talks1
Reflection AI1
Ricursive Intelligence1Shunya Labs1
- Zoho1