Claude Sonnet 4.5 vs Seed Oss 36B Instruct
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
Claude Sonnet 4.5 is the stronger all-rounder.Seed Oss 36B Instruct 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.
- ReasoningHard problems that need careful thinking30Claude Sonnet 4.53 of 3 tests
- FactsGetting facts right instead of making them up20Claude Sonnet 4.52 of 2 tests
- CodingWriting and fixing software21Claude Sonnet 4.52 of 3 tests
- AgentsCarrying out multi-step tasks on its own10Claude Sonnet 4.51 of 1 test
- Long documentsFinding answers in very long texts10Claude Sonnet 4.51 of 1 test
- Following instructionsDoing exactly what it is asked10Claude Sonnet 4.51 of 1 test
Agents, 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.
Seed Oss 36B Instruct costs 96% 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 Claude Sonnet 4.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+37.3points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+28.8points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+15.4points ahead
Where Seed Oss 36B Instruct pulls ahead
- Recent programming contest problemsLiveCodeBench v6+3.4points 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.
CodingClaude Sonnet 4.5
- Terminal-Bench HardClaude Sonnet 4.5 by 28.835.66.8+28.8
- SciCodeClaude Sonnet 4.5 by 9.245.736.5+9.2
- LiveCodeBench v6Seed Oss 36B Instruct by 3.46467.4+3.4
ReasoningClaude Sonnet 4.5
- GPQA DiamondClaude Sonnet 4.5 by 10.883.472.6+10.8
- Humanity's Last ExamClaude Sonnet 4.5 by 7.917.89.9+7.9
- CritPtClaude Sonnet 4.5 by 1.11.10+1.1
FactsClaude Sonnet 4.5
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.5 by 37.350.913.6+37.3
- AA-Omniscience · AccuracyClaude Sonnet 4.5 by 14.732.918.2+14.7
Following instructionsClaude Sonnet 4.5
- IFBenchClaude Sonnet 4.5 by 15.457.341.9+15.4
Other results12 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-OmniscienceClaude Sonnet 4.5 by 52.4-0.1-52.5+52.4
- Artificial Analysis Coding IndexClaude Sonnet 4.5 by 35.452.116.7+35.4
- τ²-Bench Telecom (AA run)Claude Sonnet 4.5 by 28.778.149.4+28.7
- Multi-SWE-BenchClaude Sonnet 4.5 by 27.344.317+27.3
- TAU-bench (airline)Claude Sonnet 4.5 by 247046+24
- TAU-bench (retail)Claude Sonnet 4.5 by 15.886.270.4+15.8
- AIR-Bench 2024Claude Sonnet 4.5 by 14.289.875.6+14.2
- MMMLUClaude Sonnet 4.5 by 10.789.178.4+10.7
- AA Agentic IndexSeed Oss 36B Instruct by 10.217.527.7+10.2
- AA IntelligenceClaude Sonnet 4.5 by 8.620.712.1+8.6
- MMLU-ProClaude Sonnet 4.5 by 4.887.582.7+4.8
- AIME 2025Claude Sonnet 4.5 by 2.38784.7+2.3
Questions people ask
Which is better, Claude Sonnet 4.5 or Seed Oss 36B Instruct?
Claude Sonnet 4.5 wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. Seed Oss 36B Instruct wins none, but costs 96% less.
Which is better for coding?
Claude Sonnet 4.5. It wins 2 of the 3 coding tests both models report; Seed Oss 36B Instruct wins 1.
Which is cheaper?
Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; Seed Oss 36B Instruct costs $0.21 and $0.57. That makes Seed Oss 36B Instruct about 96% 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 11 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 12 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.