GLM-5 vs Sarvam 30B

Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.

Agentic
Coding
Multimodal & Grounded
Reasoning
Knowledge
Instruction Following
Multilingual
Mathematics

GLM-5· Sarvam 30B

Quick Verdict

Pick GLM-5 if you want the stronger benchmark profile. Sarvam 30B only becomes the better choice if knowledge is the priority or you want the stronger reasoning-first profile.

GLM-5 is clearly ahead on the aggregate, 75 to 48. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

GLM-5's sharpest advantage is in coding, where it averages 60.4 against 34. The single biggest benchmark swing on the page is SWE-bench Verified, 77.8% to 34%. Sarvam 30B does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.

Sarvam 30B is the reasoning model in the pair, while GLM-5 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. GLM-5 gives you the larger context window at 200K, compared with 64K for Sarvam 30B.

Operational tradeoffs

PriceFree*Free*
Speed74 t/sN/A
TTFT1.64sN/A
Context200K64K

Decision framing

BenchLM keeps the benchmark table and the operator tradeoffs on the same page so a better score does not hide a materially slower, pricier, or smaller-context model.

Runtime metrics show N/A when BenchLM does not have a sourced snapshot for that exact model. The scoring rules and freshness policy are documented on the methodology page.

BenchmarkGLM-5Sarvam 30B
AgenticGLM-5 wins
Terminal-Bench 2.056.2%
BrowseComp62%35.5%
OSWorld-Verified58%
Claw-Eval57.7%
QwenClawBench54.1%
QwenWebBench1315
TAU3-Bench65.6%
VITA-Bench37.0%
DeepPlanning14.6%
Toolathlon38%
MCP Atlas31.1%
MCP-Tasks60.8%
WideResearch69.8%
Tau2-Airline80.5%
Tau2-Telecom98.2%
BFCL v470.8%
CodingGLM-5 wins
HumanEval80%92.1%
SWE-bench Verified77.8%34%
SWE-bench Verified*72.8%
LiveCodeBench52%
LiveCodeBench v685.6%70.0%
SWE-bench Pro55.1%
SWE Multilingual73.3%
NL2Repo35.9%
SWE-Rebench62.8%
React Native Evals74.2%
Multimodal & Grounded
MMMU-Pro66%
OfficeQA Pro73%
Reasoning
MuSR82%
BBH83%
LongBench v260.8%
MRCRv273%
AI-Needle63.3%
gpqaDiamond66.5%
KnowledgeSarvam 30B wins
MMLU91.7%85.1%
GPQA86%
GPQA-D81.6%
SuperGPQA66.8%
MMLU-Pro85.7%80%
MMLU-Pro (Arcee)85.8%
MMLU-Redux94.4%
C-Eval92.8%
HLE27.2%
FrontierScience74%
SimpleQA84%
Instruction Following
IFEval92.6%
IFBench72.3%
Multilingual
MGSM84%
MMLU-ProX83.1%
NOVA-6355.1%
INCLUDE84.9%
PolyMath65.2%
VWT2k-lite82.1%
MAXIFE85.6%
MathematicsGLM-5 wins
AIME 202388%
AIME 202490%
AIME 202593.3%80%
AIME2695.8%
AIME25 (Arcee)93.3%
HMMT Feb 202384%
HMMT Feb 202486%
HMMT Feb 202585%
HMMT Feb 202597.5%73.3%
HMMT Nov 202596.9%74.2%
HMMT Feb 202686.4%
MMAnswerBench82.5%
BRUMO 202587%
MATH-50097.4%97%
Frequently Asked Questions (5)

Which is better, GLM-5 or Sarvam 30B?

GLM-5 is ahead overall, 75 to 48. The biggest single separator in this matchup is SWE-bench Verified, where the scores are 77.8% and 34%.

Which is better for knowledge tasks, GLM-5 or Sarvam 30B?

Sarvam 30B has the edge for knowledge tasks in this comparison, averaging 80 versus 67.7. Inside this category, MMLU is the benchmark that creates the most daylight between them.

Which is better for coding, GLM-5 or Sarvam 30B?

GLM-5 has the edge for coding in this comparison, averaging 60.4 versus 34. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.

Which is better for math, GLM-5 or Sarvam 30B?

GLM-5 has the edge for math in this comparison, averaging 92.1 versus 86.5. Inside this category, HMMT Feb 2025 is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, GLM-5 or Sarvam 30B?

GLM-5 has the edge for agentic tasks in this comparison, averaging 58.3 versus 35.5. Inside this category, BrowseComp is the benchmark that creates the most daylight between them.

Last updated: April 3, 2026

Weekly LLM Benchmark Digest

Get notified when new models drop, benchmark scores change, or the leaderboard shifts. One email per week.

Free. No spam. Unsubscribe anytime. We only store derived location metadata for consent routing.