Kimi K2 Instruct vs Sarvam 105B
Wins 4 of 6 areas
Coding · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
Kimi K2 Instruct is the stronger all-rounder.Sarvam 105B is cheaper.
Scores updated · 23 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software31Kimi K2 Instruct3 of 4 tests
- FactsGetting facts right instead of making them up20Kimi K2 Instruct2 of 2 tests
- Long documentsFinding answers in very long texts10Kimi K2 Instruct1 of 1 test
- Following instructionsDoing exactly what it is asked10Kimi K2 Instruct1 of 1 test
- AgentsCarrying out multi-step tasks on its own11Even1 each
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Sarvam 105B costs 93% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Kimi K2 Instruct pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+23.7points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+22points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+20.8points ahead
Where Sarvam 105B pulls ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+35.4points ahead
- Recent programming contest problemsLiveCodeBench v6+18points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+4.6points ahead
Every test, side by side
All 23 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K2 Instruct
- Terminal-Bench HardKimi K2 Instruct by 2223.51.5+22
- SWE-bench VerifiedKimi K2 Instruct by 20.865.845+20.8
- LiveCodeBench v6Sarvam 105B by 1853.771.7+18
- SciCodeKimi K2 Instruct by 4.330.726.4+4.3
AgentsEven
- BrowseCompSarvam 105B by 35.414.149.5+35.4
- GDPValKimi K2 Instruct by 6.318.211.9+6.3
ReasoningEven
- Humanity's Last ExamSarvam 105B by 4.66.411+4.6
- GPQA DiamondKimi K2 Instruct by 2.976.773.8+2.9
- CritPttie00tie
FactsKimi K2 Instruct
- AA-Omniscience · Non-hallucinationKimi K2 Instruct by 23.730.36.6+23.7
- AA-Omniscience · AccuracyKimi K2 Instruct by 7.825.417.6+7.8
Following instructionsKimi K2 Instruct
- IFBenchKimi K2 Instruct by 7.341.734.4+7.3
Other results10 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AIME 2025Sarvam 105B by 47.249.596.7+47.2
- HMMT 2025Sarvam 105B by 4738.885.8+47
- AA-OmniscienceKimi K2 Instruct by 32.8-26.6-59.4+32.8
- τ²-Bench Telecom (AA run)Kimi K2 Instruct by 26.673.446.8+26.6
- Artificial Analysis Coding IndexKimi K2 Instruct by 16.125.99.8+16.1
- AA Agentic IndexKimi K2 Instruct by 1337.724.7+13
- AA IntelligenceKimi K2 Instruct by 6.515.38.8+6.5
- IFEvalKimi K2 Instruct by 589.884.8+5
- MATH-500 (EM)Sarvam 105B by 1.297.498.6+1.2
- MMLU-Protie81.181.7tie
Questions people ask
Which is better, Kimi K2 Instruct or Sarvam 105B?
Kimi K2 Instruct wins four of the six areas where both have results: coding, facts, long documents and following instructions. Sarvam 105B wins none, but costs 93% less. They are level on agents and reasoning.
Which is better for coding?
Kimi K2 Instruct. It wins 3 of the 4 coding tests both models report; Sarvam 105B wins 1.
Which is cheaper?
Kimi K2 Instruct costs $0.60 per million input tokens and $2.50 per million output tokens; Sarvam 105B costs $0.04 and $0.17. That makes Sarvam 105B about 93% cheaper for the same work.
How do you compare the two?
We use the 23 benchmark tests both models have published scores on. The verdict counts the 13 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 10 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.