Indian AI company Sarvam released Sarvam Prism on November 3, a "System One" decision model for enterprise workflows: given a context and questions with fixed answer options, it returns a probability for every option in a single forward pass instead of generating text. Sarvam says that across the 25 decision models it tested, Prism averages 71.8 over three indices with a median response time of 44 ms on one B200, the highest score under 100 ms, and scores 82.5 on its Indian-language benchmark. Weights are on Hugging Face under Apache 2.0, with a hosted API as well. All scores are Sarvam's own measurements.
Key Takeaways
- ✓Trained on the Sarvam 105B (105B-10B MoE) base; it only outputs option probabilities, so the answer is always one of the given options.
- ✓Vendor-reported indices: Jev Decision Index 67.0, Sarvam's recreation of the Artificial Analysis decision index 65.9, internal Indian-language benchmark 82.5, average 71.8; Perplexity Decider v1.2 averages 68.0 at about 117 ms.
- ✓Calibration: on English answers at 90%+ confidence it expects 3.3% errors and makes 2.2%; in Indian languages 5.4% expected vs 4.4% actual, while rivals underestimate errors by 1.5x to 4.8x.
- ✓By sorting options into a fixed order before scoring, it changed 0 of 2,000 answers across four shuffled option orders.
- ✓On 700 recorded agent runs it judges correctly 78.5% of the time overall, vs 70.7% for Claude Haiku 5.5.
- ✓Self-hosting needs one FP8-capable GPU with about 120 GB free memory (such as B200 or H200) and about 106 GB of disk for weights.
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
A lot of enterprise work is a chain of small decisions: which team gets a ticket, whether a run needs human review, whether an agent run succeeded, whether a prompt contains an injection. General chat models are slow and costly for this, may answer outside the allowed options, and often have unreliable confidence. Sarvam notes that Typeface created the "System One decision model" category last month with its Jev model; Prism is Sarvam's first model in this category and its latest open-weight release after open-sourcing Sarvam 30B and 105B in March.
How it works
Prism is trained on the Sarvam 105B-10B MoE base. You pass a context state and a set of questions (choice, yes/no, or score), and it returns probabilities for every option in one forward pass. To remove option-order bias, it sorts options into a fixed order before scoring and maps probabilities back to the original names.
Benchmarks
All numbers are Sarvam's own: among the 25 decision models it tested, Prism averages 71.8 over three indices at a 44 ms median response time (one B200, 1,000 sequential requests). Perplexity Decider v1.2 scores 68.0 at 117 ms, Jev 1.14 64.5 (196 ms reported) and Microsoft-Decision-1 61.1 (85 ms reported); Sarvam measured those two API-only models at 431 ms and 533 ms from its cluster. Sarvam acknowledges a few models score higher overall but take two to six times as long. The second index is Sarvam's own recreation of the Artificial Analysis decision index, not an official Artificial Analysis result, and the third is an internal Indian-language benchmark.
Impact for developers and enterprises
Teams doing routing, classification, moderation or agent-trace evaluation get a low-latency option they can self-host and fine-tune on their own data, and calibrated confidence makes threshold-based automation safer. Self-hosting needs one FP8 GPU with about 120 GB free memory; teams without one can use Sarvam's hosted API.
Editor's note
Decision models are a very new category, most comparison models are niche, and every score, latency and calibration figure comes from Sarvam with no independent replication yet. Benchmark it on your own routing or classification tasks before switching.
Sources: Sarvam on X, Introducing Sarvam Prism, Hugging Face model card.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.