DeepSeek-R1 0528 vs Sarvam 105B
Wins 5 of 6 areas
Coding · Reasoning · Facts · Long documents · Following instructions
Wins 1 of 6 areas
Agents
DeepSeek-R1 0528 is the stronger all-rounder.Sarvam 105B is cheaper and better at agents.
Scores updated · 21 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.
- ReasoningHard problems that need careful thinking30DeepSeek-R1 05283 of 3 tests
- CodingWriting and fixing software20DeepSeek-R1 05282 of 3 tests · 1 tie
- FactsGetting facts right instead of making them up20DeepSeek-R1 05282 of 2 tests
- Long documentsFinding answers in very long texts10DeepSeek-R1 05281 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-R1 05281 of 1 test
- AgentsCarrying out multi-step tasks on its own02Sarvam 105B2 of 2 tests
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 95% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where DeepSeek-R1 0528 pulls ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+14.4points ahead
- Code for real scientific research problemsSciCode+13.9points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+12.9points ahead
Where Sarvam 105B pulls ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+40.6points ahead
- Real work tasks from 44 professionsGDPVal+2.9points ahead
Every test, side by side
All 21 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-R1 0528
- Terminal-Bench HardDeepSeek-R1 0528 by 14.415.91.5+14.4
- SciCodeDeepSeek-R1 0528 by 13.940.326.4+13.9
- SWE-bench Verifiedtie44.645tie
AgentsSarvam 105B
- BrowseCompSarvam 105B by 40.68.949.5+40.6
- GDPValSarvam 105B by 2.9911.9+2.9
ReasoningDeepSeek-R1 0528
- GPQA DiamondDeepSeek-R1 0528 by 7.581.373.8+7.5
- Humanity's Last ExamDeepSeek-R1 0528 by 4.815.811+4.8
- CritPtDeepSeek-R1 0528 by 1.41.40+1.4
FactsDeepSeek-R1 0528
- AA-Omniscience · AccuracyDeepSeek-R1 0528 by 12.930.517.6+12.9
- AA-Omniscience · Non-hallucinationDeepSeek-R1 0528 by 1016.66.6+10
Following instructionsDeepSeek-R1 0528
- IFBenchDeepSeek-R1 0528 by 5.239.634.4+5.2
Other results9 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.
- AA-OmniscienceDeepSeek-R1 0528 by 32-27.4-59.4+32
- Artificial Analysis Coding IndexDeepSeek-R1 0528 by 14.2249.8+14.2
- τ²-Bench Telecom (AA run)Sarvam 105B by 10.336.546.8+10.3
- AIME 2025Sarvam 105B by 9.287.596.7+9.2
- HMMT 2025Sarvam 105B by 6.479.485.8+6.4
- AA IntelligenceDeepSeek-R1 0528 by 4.313.18.8+4.3
- AA Agentic IndexSarvam 105B by 3.920.824.7+3.9
- MMLU-ProDeepSeek-R1 0528 by 3.38581.7+3.3
- MATH-500 (EM)tie9898.6tie
Questions people ask
Which is better, DeepSeek-R1 0528 or Sarvam 105B?
DeepSeek-R1 0528 wins five of the six areas where both have results: coding, reasoning, facts, long documents and following instructions. Sarvam 105B wins agents, and costs 95% less.
Which is better for coding?
DeepSeek-R1 0528. It wins 2 of the 3 coding tests both models report; Sarvam 105B wins none, and 1 is a tie.
Which is cheaper?
DeepSeek-R1 0528 costs $1.35 per million input tokens and $3.00 per million output tokens; Sarvam 105B costs $0.04 and $0.17. That makes Sarvam 105B about 95% cheaper for the same work.
How do you compare the two?
We use the 21 benchmark tests both models have published scores on. The verdict counts the 12 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 9 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.