Sarvam AILanguage model
Sarvam 30B
Sarvam 30B has no overall index value yet because there are too few independent benchmark results.
PopularityRank 339 of 785
38out of 100same as last month
based on search, usage and attention
Summary
Sarvam 30B is a sparse Mixture-of-Experts model with 2.4 billion non-embedding active parameters. Sarvam targets practical deployment, Indian languages, coding, reasoning, conversation, and tool calls. The card also highlights multilingual voice calls. Grouped Query Attention reduces KV-cache pressure, while the 128,000-token window combines long inputs with a focus on more constrained inference environments.
Sarvam’s official release post describes 30B as a production model and documents API access plus downloadable weights. Sarvam 105B follows the same open-model direction at larger scale, but it is a different MoE design. Sarvam documents optimizations for servers, L40S, and Apple Silicon; the card also highlights 22 Indian languages.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 93 published benchmark results
Page 1 of 8
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
CritPt v2: Overall (Show measurement, test conditions, and source)
- Source value
- 0.28571428571428603
- Score
- 0.28571428571428603
- Metric
- pass@1 accuracy
- Unit
- %
- Benchmark version
- v2
- Category
- reasoning
- Task
- overall
- Task label
- Overall
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- disputed
- Retrieved at
- 2026-11-02
- Methodology
- Exact AA configuration: Sarvam 30B (High) (release sarvam-30b, row sarvam-30b, UUID 764dd095-6ebd-4760-8eb3-cbc40964db2e, reasoning=true, effort=high, suffix=High). AA source value for average reasoning tokens before the answer: 2000. Its methodology averages 60 varied prompts and assumes 2,000 when no average is available. This is not a token cap. The full CritPt v2 run covers 70 questions with five repeats. The official grading server reports pass@1.. Source model ID: artificial-analysis/764dd095-6ebd-4760-8eb3-cbc40964db2e/sarvam-30b/sarvam-30b. Source model name: Sarvam 30B. Provider reported by source: Sarvam. Model snapshot: Sarvam 30B (High). Evaluator: Artificial Analysis. Dataset type: artificial-analysis-evaluation. Tool policy: Python functions, symbolic expressions, and numerical answers; official grading server. Attempts: 70 questions, five repeats per question, pass@1. Reasoning: reasoning enabled. Reasoning effort: high. Score display: 0.3%. Source ID: artificial-analysis. Evidence capture ID: aa-model-page-2026-11-02-67d46be20a5b. Source payload hash: 67d46be20a5b412db6cfb60ad190441fcaa5513500b8fd47d033168d0e97b6b2. Parser version: artificial-analysis-rsc-models-v3. Imported at: 2026-11-02. Source cohort: artificial-analysis-critpt-v2:v2:standalone:reasoning:effort=high:suffix=High
- Comparison methodology
- Exact AA configuration: Sarvam 30B (High) (release sarvam-30b, row sarvam-30b, UUID 764dd095-6ebd-4760-8eb3-cbc40964db2e, reasoning=true, effort=high, suffix=High). AA source value for average reasoning tokens before the answer: 2000. Its methodology averages 60 varied prompts and assumes 2,000 when no average is available. This is not a token cap. The full CritPt v2 run covers 70 questions with five repeats. The official grading server reports pass@1.. Tool policy: Python functions, symbolic expressions, and numerical answers; official grading server. Attempts: 70 questions, five repeats per question, pass@1. Reasoning: reasoning enabled. Reasoning effort: high
- Context
- This AA source field contains the original CritPt score. The earlier CritPt v2 attribution was incorrect; the value and former label remain as disputed historical evidence.. The source dataset marks this result as disputed. It remains visible as documented evidence, but is not reliable comparative proof.
IFBench: Overall (Show measurement, test conditions, and source)
- Source value
- 26.462585034013603
- Score
- 26.462585034013603
- Metric
- IFBench prompt-level loose accuracy (pass@1)
- Unit
- %
- Benchmark version
- AA test set: 294 questions
- Category
- general
- Task
- overall
- Task label
- Overall
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- superseded
- Retrieved at
- 2026-11-02
- Methodology
- Exact AA configuration: Sarvam 30B (High) (release sarvam-30b, row sarvam-30b, UUID 764dd095-6ebd-4760-8eb3-cbc40964db2e, reasoning=true, effort=high, suffix=High). AA source value for average reasoning tokens before the answer: 2000. Its methodology averages 60 varied prompts and assumes 2,000 when no average is available. This is not a token cap. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.. Source model ID: artificial-analysis/764dd095-6ebd-4760-8eb3-cbc40964db2e/sarvam-30b/sarvam-30b. Source model name: Sarvam 30B. Provider reported by source: Sarvam. Model snapshot: Sarvam 30B (High). Evaluator: Artificial Analysis. Dataset type: artificial-analysis-evaluation. Tool policy: Single-turn answer evaluated with the official IFBench source code. Attempts: pass@1, 5 repeats per question. Reasoning: reasoning enabled. Reasoning effort: high. Score display: 26.5%. Source ID: artificial-analysis. Evidence capture ID: aa-model-page-2026-11-02-67d46be20a5b. Source payload hash: 67d46be20a5b412db6cfb60ad190441fcaa5513500b8fd47d033168d0e97b6b2. Parser version: artificial-analysis-rsc-models-v3. Imported at: 2026-11-02. Source cohort: artificial-analysis-ifbench:AA test set: 294 questions:standalone:reasoning:effort=high:suffix=High
- Comparison methodology
- Exact AA configuration: Sarvam 30B (High) (release sarvam-30b, row sarvam-30b, UUID 764dd095-6ebd-4760-8eb3-cbc40964db2e, reasoning=true, effort=high, suffix=High). AA source value for average reasoning tokens before the answer: 2000. Its methodology averages 60 varied prompts and assumes 2,000 when no average is available. This is not a token cap. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.. Tool policy: Single-turn answer evaluated with the official IFBench source code. Attempts: pass@1, 5 repeats per question. Reasoning: reasoning enabled. Reasoning effort: high
- Context
- The often cited 58 refers to verifiable out-of-domain constraint types, not the number of questions in AA's test run.. This result is retained as historical evidence and is not interchangeable with current results.
IFBench: Overall (Show measurement, test conditions, and source)
- Source value
- 26.462585034013603
- Score
- 26.462585034013603
- Metric
- IFBench prompt-level loose accuracy (pass@1)
- Unit
- %
- Benchmark version
- AA test set: 294 questions
- Category
- general
- Task
- overall
- Task label
- Overall
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Exact AA configuration: Sarvam 30B (High) (release sarvam-30b, row sarvam-30b, UUID 764dd095-6ebd-4760-8eb3-cbc40964db2e, reasoning=true, effort=high, suffix=High). This source capture does not provide an average reasoning-token count. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.. Source model ID: artificial-analysis/764dd095-6ebd-4760-8eb3-cbc40964db2e/sarvam-30b/sarvam-30b. Source model name: Sarvam 30B. Provider reported by source: Sarvam. Model snapshot: Sarvam 30B (High). Evaluator: Artificial Analysis. Dataset type: artificial-analysis-evaluation. Tool policy: Single-turn answer evaluated with the official IFBench source code. Attempts: pass@1, 5 repeats per question. Reasoning: reasoning enabled. Reasoning effort: high. Score display: 26.5%. Source ID: artificial-analysis. Evidence capture ID: aa-model-page-2026-11-07-b96e3ffc8c5b. Source payload hash: b96e3ffc8c5b052aa56f96d05e89d02b8cc234a75085820fca0ca2a39e498274. Parser version: artificial-analysis-rsc-models-v4. Imported at: 2026-11-07. Source cohort: artificial-analysis-ifbench:AA test set: 294 questions:standalone:reasoning:effort=high:suffix=High
- Comparison methodology
- Exact AA configuration: Sarvam 30B (High) (release sarvam-30b, row sarvam-30b, UUID 764dd095-6ebd-4760-8eb3-cbc40964db2e, reasoning=true, effort=high, suffix=High). This source capture does not provide an average reasoning-token count. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.. Tool policy: Single-turn answer evaluated with the official IFBench source code. Attempts: pass@1, 5 repeats per question. Reasoning: reasoning enabled. Reasoning effort: high
- Context
- The often cited 58 refers to verifiable out-of-domain constraint types, not the number of questions in AA's test run.
IndicQuest v2 LLM-as-a-judge weighted accuracy (English) (Show measurement, test conditions, and source)
- Source value
- 48.7
- Score
- 48.7
- Metric
- LLM-as-a-judge weighted accuracy for English
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Sample
- 3471
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 5, exact Sarvam 30B row for the English subset; the paper states 3,471 English question-answer pairs and says the same 3,471 responses per model are used for deterministic metrics. LLM-as-a-judge uses Gemma 3 12B. The source does not publish a confidence interval. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 exact-substring accuracy (English) (Show measurement, test conditions, and source)
- Source value
- 27.7
- Score
- 27.7
- Metric
- Exact-substring accuracy for English
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Sample
- 3471
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 5, exact Sarvam 30B row for the English subset; the paper states 3,471 English question-answer pairs and says the same 3,471 responses per model are used for deterministic metrics. LLM-as-a-judge uses Gemma 3 12B. The source does not publish a confidence interval. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 word-overlap accuracy (English) (Show measurement, test conditions, and source)
- Source value
- 27.9
- Score
- 27.9
- Metric
- Word-overlap accuracy for English
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Sample
- 3471
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 5, exact Sarvam 30B row for the English subset; the paper states 3,471 English question-answer pairs and says the same 3,471 responses per model are used for deterministic metrics. LLM-as-a-judge uses Gemma 3 12B. The source does not publish a confidence interval. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 LLM-as-a-judge weighted accuracy (Gujarati) (Show measurement, test conditions, and source)
- Source value
- 40.7
- Score
- 40.7
- Metric
- LLM-as-a-judge weighted accuracy for Gujarati
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 6, exact “Weighted” row for Sarvam 30B and the named language column. This is domain-size-weighted LLM-as-a-judge accuracy; the paper identifies Gemma 3 12B as judge. It reports 69,420 multilingual question-answer pairs across 20 languages but gives no per-language denominator next to these Table 6 cells. No confidence interval is published. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 LLM-as-a-judge weighted accuracy (Hindi) (Show measurement, test conditions, and source)
- Source value
- 38.6
- Score
- 38.6
- Metric
- LLM-as-a-judge weighted accuracy for Hindi
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 6, exact “Weighted” row for Sarvam 30B and the named language column. This is domain-size-weighted LLM-as-a-judge accuracy; the paper identifies Gemma 3 12B as judge. It reports 69,420 multilingual question-answer pairs across 20 languages but gives no per-language denominator next to these Table 6 cells. No confidence interval is published. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 LLM-as-a-judge weighted accuracy (Assamese) (Show measurement, test conditions, and source)
- Source value
- 38.4
- Score
- 38.4
- Metric
- LLM-as-a-judge weighted accuracy for Assamese
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 6, exact “Weighted” row for Sarvam 30B and the named language column. This is domain-size-weighted LLM-as-a-judge accuracy; the paper identifies Gemma 3 12B as judge. It reports 69,420 multilingual question-answer pairs across 20 languages but gives no per-language denominator next to these Table 6 cells. No confidence interval is published. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 LLM-as-a-judge weighted accuracy (Telugu) (Show measurement, test conditions, and source)
- Source value
- 37.6
- Score
- 37.6
- Metric
- LLM-as-a-judge weighted accuracy for Telugu
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 6, exact “Weighted” row for Sarvam 30B and the named language column. This is domain-size-weighted LLM-as-a-judge accuracy; the paper identifies Gemma 3 12B as judge. It reports 69,420 multilingual question-answer pairs across 20 languages but gives no per-language denominator next to these Table 6 cells. No confidence interval is published. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 LLM-as-a-judge weighted accuracy (Odia) (Show measurement, test conditions, and source)
- Source value
- 37.4
- Score
- 37.4
- Metric
- LLM-as-a-judge weighted accuracy for Odia
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 6, exact “Weighted” row for Sarvam 30B and the named language column. This is domain-size-weighted LLM-as-a-judge accuracy; the paper identifies Gemma 3 12B as judge. It reports 69,420 multilingual question-answer pairs across 20 languages but gives no per-language denominator next to these Table 6 cells. No confidence interval is published. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
IndicQuest v2 LLM-as-a-judge weighted accuracy (Sanskrit) (Show measurement, test conditions, and source)
- Source value
- 35.3
- Score
- 35.3
- Metric
- LLM-as-a-judge weighted accuracy for Sanskrit
- Unit
- %
- Benchmark version
- L3Cube-IndicQuest v2, arXiv:2608.15535v1
- Category
- factual-knowledge
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Table 6, exact “Weighted” row for Sarvam 30B and the named language column. This is domain-size-weighted LLM-as-a-judge accuracy; the paper identifies Gemma 3 12B as judge. It reports 69,420 multilingual question-answer pairs across 20 languages but gives no per-language denominator next to these Table 6 cells. No confidence interval is published. The paper’s exact row label is “Sarvam 30B”; the matching official public model card identifies repository `sarvamai/sarvam-30b`. The paper does not bind its evaluation to an immutable repository revision or API route, and it does not publish decoding controls for this run.
- Context
- The paper labels the system only “Sarvam 30B”. The matching model card supports the family/repository mapping, not the tested commit or endpoint. The nine non-English Table 6 cells have no explicitly stated per-language N; neither N nor confidence intervals are inferred.
Page 1 of 8
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
Sarvam 30BDeepSeek-R1 0528 38Ministral 3 14B 38Motif 3 38GPT-4.1 38
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Sarvam 30B, chosen by model line, developer, size, price and release date.
- Sarvam-MSarvam AI · same developer
- Sarvam 105BSarvam AI · same developer
- Nemotron 3 NanoNVIDIA · similar size
- Nemotron Cascade 2 30B A3BNVIDIA · similar size
- Qwen 3.5 27BAlibaba · similar size
- GLM-4.7-FlashZ.ai · similar size
- Nemotron 3 Nano Omni 30B A3B ReasoningNVIDIA · similar size
- Qwen 3.6 27BAlibaba · similar size
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about Sarvam 30B
How large is the context window of Sarvam 30B?
128,000 tokens. That is how much text Sarvam 30B reads at once, about 190 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
Sarvam 30BRank 98 of 103128K
DeepSeek-V3.2128K
DeepSeek-R1128K
Can I run Sarvam 30B locally?
Yes, the weights are open.
Stated by Sarvam AI
- Open weights
- yes
- License
- Apache 2.0
- Parameters
- 30B
- active per token
- n/a
Sources and data date
Show all 18 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
| Evaluator | Test series | Composite indices | Recorded active measurements |
|---|---|---|---|
| Artificial Analysis | 1 | 0 | 1 |
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 27, 2026. |
| Output | Documented modalities. Retrieved October 27, 2026. |
| Research status | Research date October 15, 2026. 9 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | score-units-and-evaluation-details Note The card leaves table score units, most benchmark versions, task dates, sampling and prompting unspecified. Tool-use parenthetical alternatives were excluded because their mapping is not clear. |
| Unresolved | aime-value-and-evaluation-methodology Note The official release page's Sarvam-30B table gives 80.0 (96.7), while the accompanying prose says 88.3 Pass@1 and 96.7 with tool use. Its 88.3 prose value matches the Sarvam-105B table value, but the publisher provides no correction. Evaluation date, sample count and most harness settings are also absent. |
| Not published | tested-checkpoint-or-api-revision Note The paper names Sarvam 30B and the matching model card identifies the public repository, but no immutable tested commit or API route is reported. |
| Not published | per-language-sample-count-table-6 Note Table 6 does not report a per-language N for the nine Indic-language cells. The reported total of 69,420 pairs across 20 languages is not used to infer a per-cell denominator. |
| Not published | confidence-interval-95 Note The paper does not provide confidence intervals for the reported accuracy cells. |
| Additional source | Hugging Face (retrieved October 15, 2026; October 17, 2026) · Sarvam 30B model card · Editorial description reviewed October 2, 2026 |
| Additional source | Artificial Analysis (retrieved November 2, 2026) |
| Additional source | Artificial Analysis (retrieved November 2, 2026; November 7, 2026) |
| Additional source | Sarvam: Open-Sourcing Sarvam 30B and 105B (retrieved October 17, 2026; October 22, 2026; October 23, 2026) · Sarvam 30B and 105B release post · Editorial description reviewed October 2, 2026 |
| Additional source | L3Cube-IndicQuest v2, arXiv:2608.15535v1, Tables 5 and 6 (retrieved October 17, 2026) |
| Additional source | XL-SafetyBench, arXiv:2605.05662v2 (retrieved October 18, 2026) |
| Additional source | AI4Bharat / IIT Madras and NVIDIA, Indic-LLM-Eval-Suite (retrieved October 18, 2026) |
| Additional source | IndicEval (Aviral Kaintura) (retrieved October 19, 2026) |
| Additional source | Sthanika, MILU 2026 rerun (retrieved October 21, 2026) |
| Additional source | Goedel Machines, Sarvam's illusion of safety (retrieved October 21, 2026; October 23, 2026) |
| Additional source | Model metadata (retrieved October 19, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |