osmAPI

Wishing our community of developers, builders, and founders a peaceful and meaningful Good Friday. 🕊️ Today is a reminder that the strongest foundations are not built overnight— They are shaped through patience, precision, and purpose. At osmAPI, we believe the most impactful systems are the ones that work quietly, reliably, and consistently in the background—empowering you to focus on what truly matters: building, innovating, and growing. As you take this time to reflect and recharge, know that every great creation begins with a strong and steady foundation. Here’s to building with clarity, strength, and intention. #goodfriday #developers #tech #openAI #API

Incredible depth in this breakdown 👏 We recently worked on Sarvam-30B from a different angle — optimizing it for real-world deployment. ~3.5x compression using TurboQuant with just ~2.5% drop. 64GB → 18.6GB. Sarvam AI is pushing model capabilities — we’re focused on making them practical at scale 👇 x.com/junafinity/sta…

The first Indian open source model trained from scratch, Sarvam, 30B and 105B is really good. The 105B one is head to head with Deepseek R1 when it was released. Apache 2.0 license - uses a mixture-of-experts (MoE) architecture. 105 billion total parameters but only activates 9 billion per token. - Support for all 22 official Indian languages (e.g., Hindi, Tamil, Bengali, Punjabi) plus English, with strong handling of code-mixed inputs like Hinglish. Optimized for voice-first interactions, including multimodal capabilities (text-to-speech, speech-to-text, and document vision models released alongside). - Focus on reasoning, agentic tasks (e.g., tool use, web browsing), mathematics, programming, and STEM domains. - Pre-training: Sarvam 30B was trained on ~16 trillion tokens; Sarvam 105B on ~12 trillion tokens (a counterintuitive but reported split, likely due to efficiency optimizations). Datasets were curated in-house: ~15-20% Indian-origin data, including code, web content, specialized knowledge (e.g., math, STEM), and multilingual text balanced across Indian languages. --- Sarvam AI the company behind Sarvam, is an Indian startup based in Bengaluru, did release earlier open-source models trained from scratch, such as Sarvam 2B in December 2025 (a 2-billion-parameter model pre-trained on 4 trillion tokens).

@pratykumar Huge move by Sarvam AI 👏 We recently optimized Sarvam-30B — compressing it ~3.5x using TurboQuant while retaining ~97.5% performance. 64GB → 18.6GB. Open models are powerful — making them lightweight is what unlocks real-world adoption 👇 x.com/junafinity/sta…

📢 Open-sourcing the Sarvam 30B and 105B models! Trained from scratch with all data, model research and inference optimisation done in-house, these models punch above their weight in most global benchmarks plus excel in Indian languages. Get the weights at Hugging Face and AIKosh. Thanks to the good folks at SGLang for day 0 support, vLLM support coming soon. Links, benchmark scores, examples, and more in our blog - sarvam.ai/blogs/sarvam-3…

Incredible to see India building at this level 🇮🇳 We recently optimized Sarvam-30B — compressing it ~3.5x using TurboQuant while retaining ~97.5% performance. 64GB → 18.6GB. Sarvam AI is building powerful models — we’re focused on making them usable at scale 👇 x.com/junafinity/sta…

Homegrown Models Power India’s Leap! During the India AI Impact Summit, experts highlighted how India went beyond hosting and demonstrated real innovation with the launch of open-source models Sarvam-30B and Sarvam-105B, built from the ground up to support developers, enterprises and government with a strong domestic alternative. From the multilingual “Vikram” chatbot to the upcoming SARVAM KAZE wearable device, the showcase reflected India’s commitment to democratising cutting-edge technology designed and built in the country. #IndiaAIImpactSummit2026

@himanshustwts @SarvamAI 30B + 105B with strong benchmarks 👀 What’s equally exciting is how much these models can be optimized post-training. We shrunk Sarvam-30B ~3.5x with minimal loss using TurboQuant. Scale + efficiency is the real game 👇 x.com/junafinity/sta…

sarvam just announced 30B and 105B models! 30B: + pre-trained on 16T tokens + context length of 32k tokens. 120B: + context length of 128k tokens + on-par with most other open closed source ones of its class benchmarks below!

@SarvamForDevs Huge step by Sarvam with FP8 🔥 We explored a different direction with Sarvam-30B — compressing it ~3.5x using TurboQuant while retaining ~97.5% performance. From 64GB → 18.6GB. Same intelligence, far lower cost. Check here for details - x.com/junafinity/sta…

Releasing Sarvam-30B & Sarvam-105B in FP8 FP8 quantization via ModelOpt PTQ. Weights quantized to FP8, KV cache scales calibrated so the serving engine stores KV in FP8 at runtime. ~50% memory reduction vs BF16, meaningfully faster inference, minimal quality loss. Available on HuggingFace: Sarvam-30B FP8: huggingface.co/sarvamai/sarva… Sarvam-105B FP8: huggingface.co/sarvamai/sarva…

Hi @SarvamAI @SarvamForDevs We used TurboQuant to shrink Sarvam30B ~3.5 times smaller while losing only 2.5% performance compared to BF16

Using @SarvamAI @SarvamForDevs we've created the world's first native Indic Agent

@M12vc, — led by @abhivoice — have been actively backing next-generation AI, SaaS, fintech, deep-tech, and startup ecosystem platforms across India and global markets. From @arizeai and @datologyai to @Workday , @builderio , and Edera- The portfolio reflects founders building product-led, AI-powered companies at scale across enterprise software, financial services, consumer tech, and developer ecosystems. As AI becomes foundational to innovation across these sectors, managing multi-LLM environments, optimizing infrastructure spend, benchmarking model performance, and ensuring secure deployment are becoming critical for scaling efficiently. @osmAPI_off is a unified AI orchestration layer built for AI-first companies — enabling seamless integration of multiple LLM providers through a single API, real-time monitoring of token usage and costs, cross-model benchmarking, and flexible deployment. It helps teams simplify AI infrastructure while maintaining visibility, control, and scalability. We invite you and your portfolio companies to try osmAPI and evaluate how streamlined AI orchestration can accelerate experimentation, improve cost efficiency, and support sustainable, secure AI growth. @JLopatecki , @aparnadhinak , @arimorcos , @hurrycane , @Levitt_Matt ,@JerryHTing , @Steve8708 , @bdlocks

@InfoEdgeVC , led by @sbikh with partners @kitty_agar and @Chinmay Agarwal, has consistently backed category-leading and technology-driven businesses across AI, consumer internet, fintech, edtech, foodtech, and deep tech. From @GnaniAi , 30Sundays and ShortTok to @aftershootco , @Shipsy_AI , @OsfinAI , and @DviewTech — the ecosystem reflects strong conviction in AI-first and scalable digital platforms. As AI adoption accelerates across neuro AI, video intelligence, geospatial systems, BCI, fintech, and consumer platforms, managing multiple LLM providers, infrastructure costs, performance benchmarks, and deployment flexibility is becoming increasingly complex and mission-critical. @osmAPI_off acts as a unified AI orchestration layer — enabling teams to integrate multiple LLMs through a single API, monitor token usage and spending in real time, benchmark model performance, and deploy flexibly. It simplifies AI infrastructure so product teams can focus on innovation instead of backend complexity. We would be excited for you and your portfolio founders to explore osmAPI and evaluate how streamlined AI orchestration can improve scalability, governance, and cost efficiency across AI-driven businesses. @ganeshgo , @Ananth_Nagaraj , @bharaths1987, @harshithdwivedi, @goenka01 , @jainendra_batra , @ankit_bhati1, @SohamChokshi , @dhr00v

@GembaCapital , led by @Adithpodhar and Govind Agarwal , has consistently backed early-stage, technology-first companies across fintech, insuretech, proptech, agri-tech, field service automation, and D2C platforms. From @ai_affogato and @flicktv_in to @ShowroomB2b , @AlltiusAi , @Eternz_Fashion ,@momoney_in , @Shoption1 , and @kredmint — the portfolio reflects founders building AI-enabled businesses solving operational and financial challenges at scale. As AI capabilities deepen across underwriting, property intelligence, field service optimization, investment analytics, agri supply chains, and D2C growth engines, managing multiple LLM providers, monitoring token usage, benchmarking performance, and optimizing infrastructure spend becomes increasingly critical. @osmAPI_off is designed as a unified AI orchestration layer — enabling seamless integration of multiple LLM providers through a single API, real-time tracking of token consumption and cost, cross-model benchmarking, and secure deployment options. It helps AI-first teams scale faster while maintaining cost control and governance. We would be glad for you and your portfolio founders to try osmAPI and explore how streamlined AI infrastructure can accelerate experimentation, improve efficiency, and support sustainable growth. @abhishekdua88, @MaruPelkar, @vibhanshu , @NimittJuneja , @sharad_kale_, @Vikasgarg3 , @rahulnagar87, @arthiravika18

@LightspeedIndia , led by @MohapatraHemant , @dkhare , and @bsomaia , has consistently partnered with founders building category-defining companies across AI, SaaS, fintech, conversational platforms, creator economy, and next-gen infrastructure. From @SarvamAI and @inworld_ai to @yellowdotai , @BigSurAI , @MistralAI , @eightfoldai , @SkildAI , @TheApexAi and @KodiakRobotics — the portfolio reflects bold ambition and globally scalable innovation emerging from India. As AI adoption accelerates across Indic LLMs, conversational AI, SaaS operations, fintech intelligence, and drone logistics, managing LLM performance, cost visibility, and infrastructure scalability becomes increasingly critical. @osmAPI_off acts as a unified orchestration layer for AI-first companies — enabling seamless access to multiple LLM providers through one API, real-time token usage monitoring, cross-model benchmarking, and optimized AI infrastructure spend with secure deployment options . We would be delighted for you and your portfolio companies to explore osmAPI and see how streamlined AI orchestration can accelerate product velocity while maintaining cost control and performance transparency. @vivek_raghavan , @pratykumar , @RaghuRavinutala , @KylanGibbs , @gelfenbeyn, @pathak2206 , @gupta_abhinav_ , @arthurmensch , @TimotheeLa41662 ,@GuillaumeLample , @ashutoshg, @VinodRamachand , @don_burnette ,@pazeshel

@YourNestVC , under the leadership of @skg69 , @Girishivani , and Satish Mugulavalli, has been at the forefront of backing engineering-led, IP-driven startups across robotics, 3D sensing, health AI, spacetech, telecom intelligence, and EV infrastructure. From @mikorobotusa and @cron_ai to @DozeeHealth , @AgnikulCosmos, @qpiai , @Datamotive_io , @cargofl , @contiinex , @superfoneapp , @_orbo, and @ExponentEnergy — the portfolio reflects bold founders building globally competitive technology from India. As AI adoption deepens across these sectors, managing LLM performance, cost, and scalability becomes critical. @osmAPI_off acts as a unified orchestration layer for AI-first companies — enabling teams to connect multiple LLM providers through one API. We would love for you and your portfolio companies to explore osmAPI and experience how streamlined AI infrastructure can accelerate product velocity. Strong innovation deserves equally strong AI infrastructure. @chhabratush , @MuditDand , @srinathr155 ,@Arun_Vinayak_S, @A_Yogesh_, @vijaykrishnabs, @pradeepdodle , @mSolstice, @danishjamil43 , @sinhaabhit

@scaletogether , led by @mrgirish , @manav_garg , and @shubg , has been actively backing next-generation SaaS and AI-first startups across recruitment tech, sales intelligence, privacy automation, SaaS operations, developer tools, and revenue optimization. From @recruitwithkula and @capitolai to @composio , @bytebeamio , @nexdotai , and Coreworks ai — the portfolio reflects founders building deeply product-led, AI-powered global platforms. As AI becomes foundational to SaaS innovation, managing multi-LLM environments, optimizing infrastructure spend, benchmarking model performance, and ensuring secure deployment are becoming critical for scaling efficiently. @osmAPI_off acts as a unified AI orchestration layer for AI-first companies — enabling seamless integration of multiple LLM providers through a single API, real-time monitoring of token usage and costs, cross-model benchmarking, and flexible deployment options. It simplifies AI infrastructure so teams can focus on building differentiated products rather than managing backend complexity. We would be glad for you and your portfolio founders to try osmAPI and explore how optimized AI orchestration can accelerate experimentation, improve cost visibility, and support scalable global growth. @ShaunModi, @thetalentmagnet ,@najmuzzaman , @thehusnain, @KaranVaidya6, @GanatraSoham, @kraviteza

@ElevCap , led by @mukularora , MridulArora, and @radusuma , has consistently backed high-growth, technology-first companies across fintech, generative AI, compliance, creator platforms, edtech, wealth tech, and insure tech. From @everstageinc and @atlys to @getadoptai , @FarEye , @getmaximai , @Meesho_Official , @industrybuyin_ , @sharechatapp , @IndiaMART , and Journeyfront — the portfolio reflects ambitious founders leveraging AI to transform large, complex markets. As AI becomes central to product innovation across these sectors, managing LLM performance, cost visibility, model selection, and secure scalability is increasingly strategic. @osmAPI_off is designed as a unified orchestration layer for AI-driven companies — enabling teams to integrate multiple LLM providers through a single API, monitor token usage and spend in real time, benchmark models for performance, and deploy securely . We would welcome the opportunity for you and your portfolio companies to explore osmAPI and experience how optimized AI infrastructure can accelerate experimentation, efficiency, and scale. @ankushsachdeva9 , @Deepak_anchala , @librianrahul , @AniBadam , @mohaknahta, @siva_rajamani , @vivek__s , @gk_fareye , @kushalnahata , @thisisgaurav_s, @DineshAgarwal, @vaibhavi0601 , @akshay_deo

@IndiaQuotient , led by @anandlunia , @sinhamadhukar , and @Goyal4Gagan , has built a strong track record of backing disruptive, tech-first companies across content platforms, fintech, retail, marketing automation, logistics, SME software, and HR tech. From SolaradAI and @Lendingkart to @PagarBook , @WebEngage , @fleetx_io , @VyaparApp , and @Maino_martech — the portfolio reflects founders leveraging AI and data to scale mass-market businesses across India. As AI adoption deepens across content intelligence, fintech underwriting, retail personalization, marketing automation, logistics optimization, and SME digitization, managing LLM performance, infrastructure costs, and secure deployment becomes increasingly strategic. @osmAPI_off is built as a unified orchestration layer for AI-first companies — enabling seamless integration of multiple LLM providers through a single API, real-time monitoring of token usage and spend, model benchmarking across use cases, and secure deployment . It helps teams optimize AI infrastructure while accelerating product innovation. We would be glad for you and your portfolio founders to try osmAPI and explore how streamlined AI orchestration can unlock faster experimentation, improved efficiency, and scalable growth. @rupesh_0795 , @aryaag5147 , @vinkusha , @avlesh , @abhayjeetgupta, @athearsh , @agarwalsumit12 , @Vikaskeshri1424 , @itrustdata , @abhijeetkunwar , @ravi_carpediem, @HarshLunia

@Stellaris_VP , with leaders like @alokgoyal1971 , @banglani , and @rchowdhri , has been backing high-conviction, technology-first startups across SaaS, AI, RegTech, fintech, EV infrastructure, and consumer platforms. From @whatfix and @TeamSignzy to @ArrowheadApp , @CARPLai , @dashstudioai , Drizz , @Pibit_ai , @factorsai , @GTMBuddy and @GetBoomerangAI — the portfolio showcases founders building scalable, AI-driven businesses for global markets. As AI adoption becomes central to product strategy, efficiently managing LLM performance, latency, cost, and reliability is increasingly critical. osmAPI provides a unified orchestration layer for AI-native companies — enabling seamless access to multiple LLM providers via a single API, real-time token monitoring, model benchmarking, cost optimization, and secure deployment . We would love for you and your portfolio companies to explore osmAPI and discover how streamlined AI infrastructure can strengthen product velocity and long-term scalability. @KhadimBatti , @ankitratan @lawyer_intech , @devyanigupta, @VidurMahajan1 , @VarunAlagh , @lalith0106, @Asad_abraar @shanksgan, @swamikrish2001 , @arvimurthy, @SPeddineni100, @the_geekyme

osmAPI is offering AI Compute Credits worth ₹1.5 crore exclusively for startups from Tamil Nadu to support their AI development and scaling. Tier 1 – Elite Grant • ₹5L credits for 20 high-growth AI startups (₹1 Cr total) Tier 2 – Growth Grants • ₹1L credits for 50 early-stage startups (₹50L total) Apply now: 🔗 form.startuptn.in/AICC @TheStartupTN is acting as a facilitator for this initiative. #AI #Startups #TamilNadu #AIStartups

@OfficialINDIAai @ai4bharat @SarvamAI @SarvamForDevs @caleb_friesen @Analyticsindiam

Thanks to @the_hindu for giving us a write up. Helps motivate our team to bring models from @SarvamAI @BharatGen_Com hosted in #E2E #Neysa #Yotta