Sarvam AI Launches 105B Model, Claims to Beat GPT at 20% Cost
itsOmSarraf_ · x · 2026-08-10
Indian AI company Sarvam AI announced a full-stack update at its inaugural Epoch conference, including model upgrades, an India-hosted inference layer, a self-serve agent platform, and new partnerships.
For model updates, they released new variants of Sarvam 105B tailored for conversational and work agents, the two use cases with the most traction in India. Evaluated on a production benchmark built from hundreds of millions of minutes of Indian conversations, the company claims it outperforms GPT 5.4 Mini and Gemini 3.5 Flash across voice-call capability, instruction following, tool calling, and linguistic quality.
It shows the widest gaps in instruction following and linguistic quality. Furthermore, it is priced at $0.80 per million blended tokens, a fraction of the $4.50 for GPT and $9.00 for Gemini.
More from Companies & People
- Jeff Dean Departs, DeepMind Overhauled, Gemini 3.5 Pro Cancelled: SemiAnalysis — koltregaskes · 2026-08-10
- Microsoft Execs Reading About 1873 Railroad Bubble Amid AI Capex Surge — toptickcrypto · 2026-08-10
- Traditional SaaS Under Dual Threat: AI Native Startups and Model Labs Squeezing Product Categories — signulll · 2026-08-10
- Former Li Auto ADAS chief Lang Xianpeng raises hundreds of millions in 90 days for embodied AI startup Kunlun Xing, says training costs billions — 量子位 · 2026-08-10
- AI-Generated Books Capture 40% of Self-Published Sales; Kimi K3 Sees Global Surge — FinanceYF5 · 2026-08-10
- Jeff Dean's First Talk Post-Google: How to Pick Research Problems Worth 5 Years — FinanceYF5 · 2026-08-10