Skip to content
Model Pareto

Sarvam 105B

Sarvam AIOpen weightsartificialanalysis.ai ↗
Input
$0.042/M tok
Output
$0.17/M tok
Blended
$0.074/M tok · 3:1
Speed
~95tok/sest.
Context
128Ktokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: Sarvam 105B (high). Price = AA median host price.

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.

Hardware
2× H200 141GB
FP8 weights
Throughput
~7,856 tok/s
many requests batched
Price
$0.11–0.88 /M tok
busy → light use
On your own machine
Mac Studio M5 Max 128GB (int4, ~83 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (105B params, 10.3B active per token) + 25% KV-cache headroom ≈ 131 GB
  • 2× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~7,856 output tok/s aggregate at batch 282 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~90,784 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Writing & chat

MMMLUEstimated
~last of 5 (estimated)below measured range (<81.3%)
  1. #1Gemini 3.1 Pro92.6%
  2. #2Gemma 4 31B88.4%
  3. #4gpt-oss-120b81.3%
  4. ~#5Sarvam 105BEstimated<81.3%
No published score — estimated from related benchmarks
~#46 of 57 (estimated)~1362 Elo
  1. #1Gemini 4 Argon1525
  2. #2Claude Opus 5.51507
  3. #45Trinity Large Thinking1369
  4. ~#46Sarvam 105BEstimated1362
  5. #46INTELLECT-31356
No published score — estimated from related benchmarks
~#22 of 26 (estimated)~1434 Elo
  1. #1GPT-6 Astra2173
  2. #2Claude Fable 5.12162
  3. #21Gemini 3.1 Pro1491
  4. ~#22Sarvam 105BEstimated1434
  5. #22Mistral Large 31412
No published score — estimated from related benchmarks

Hard reasoning

Agents

~#12 of 15 (estimated)~55.7%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #11Muse Glimmer65.9%
  4. ~#12Sarvam 105BEstimated55.7%
  5. #12Nemotron 3.5 Super VL NVFP452.3%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~53.8%
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. ~#3Sarvam 105BEstimated53.8%
  4. #3GPT-5.6 Terra50.2%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#19 of 22 (estimated)~50.1%
  1. #1Atria Dawn Preview92.5%
  2. #2GPT-6 Astra91.5%
  3. #18Nemotron 3.5 Bolt 74B-A9B70.0%
  4. ~#19Sarvam 105BEstimated50.1%
  5. #19Mistral Medium 3.548.6%
No published score — estimated from related benchmarks
~#51 of 92 (estimated)~823 Elo
  1. #1Claude Opus 5.51866
  2. #2Muse Wonder Full1857
  3. #50Step 3.7 Flash845
  4. ~#51Sarvam 105BEstimated823
  5. #51Qwen3.5 122B A10B802
No published score — estimated from related benchmarks
~#47 of 85 (estimated)~7.3%
  1. #1Claude Fable 5.276.4%
  2. #2Muse Wonder Full71.3%
  3. #46K2 Horizon 3.7B14.7%
  4. ~#47Sarvam 105BEstimated7.3%
  5. #47Qwen3.5 397B A17B6.9%
No published score — estimated from related benchmarks
~#61 of 91 (estimated)~11.2%
  1. #1Muse Spark 1.350.5%
  2. #2GLM-5.350.3%
  3. #60K-EXAONE 2.0 080311.5%
  4. ~#61Sarvam 105BEstimated11.2%
  5. #61Devstral Small 210.7%
No published score — estimated from related benchmarks