Inference
EN
Sarvam

Sarvam 30B

Sarvam 30B is a reasoning model from Sarvam. 14 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-03
RankedEnough results for an index.
IndexAll categories, out of 100.30.3±5.8Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.—Median 67/s
First tokenSeconds to the first output token.—Median 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.64KMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

32.15B parameters
16-bit
68.5 GB
8-bit
36.9 GB
4-bit
21.5 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 24 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.2 counted26.8206 / 218
CodingCode writing and repair.2 counted29.4198 / 209
ReasoningLogic problems and puzzles.2 counted22.4221 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted36.8216 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted13.8—
MathMath problems.3 counted37.9130 / 168

Results

14 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Evaluating Large Language Models Trained on CodeCoding92.1%4 / 22
MATH-500 Problem SetMath97.0%10 / 27
Massive Multitask Language Understanding ProfessionalKnowledge80.0%38 / 75
Massive Multitask Language UnderstandingKnowledge85.1%21 / 34
Critical Physics TasksReasoning0.3%137 / 214
Artificial Analysis Humanity's Last ExamKnowledge7.5%156 / 222
τ²-Bench Tool-Agent-User EvaluationAgentic34.5%110 / 149
Artificial Analysis GPQA DiamondKnowledge63.3%148 / 199
GPQA Diamond · BenchLMReasoning66.5%74 / 84
Harvard-MIT Mathematics Tournament February 2025Math73.3%14 / 15
Artificial Analysis IFBenchInstruction26.5%137 / 146
Artificial Analysis Long Context ReasoningReasoning0.0%204 / 217
Software Engineering Benchmark VerifiedCoding34.0%109 / 114
Harvard-MIT Mathematics Tournament November 2025Math74.2%12 / 12
BrowseCompAgentic35.5%63 / 63

Sources

BenchLMCC BY-NC 4.0

History

5 releases
Index30.6 → 30.3
23 Oct 202612 Nov 2026

Questions

How good is Sarvam 30B?
Its index is 30.3 out of 100. That is place 249 of 283 ranked models.
What is Sarvam 30B best at?
Math: 37.9 out of 100, place 130 of 168.
What is the context window of Sarvam 30B?
64K tokens.
Is Sarvam 30B open source?
Yes. Its weights are published.
When was Sarvam 30B released?
March 2026.

Badge

Sarvam 30B on Inference360: place 249 of 283, index 30.3
HTML
<a href="https://inference360.ai/models/sarvam-30b"><img src="https://inference360.ai/badges/sarvam-30b.svg" alt="Sarvam 30B on Inference360: place 249 of 283, index 30.3" height="20"></a>
Markdown
[![Sarvam 30B on Inference360: place 249 of 283, index 30.3](https://inference360.ai/badges/sarvam-30b.svg)](https://inference360.ai/models/sarvam-30b)

Compare with

LFM2.5-8B-A1B30.6DeepSeek V329.9Gemini 2.0 Flash29.5Magistral Small31.4Sarvam 105B36.7

More from Sarvam

All models
Sarvam 105B36.7