Claude Sonnet 4.6 vs Nova 2.0 Pro Preview
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Images and charts · Long documents
Wins 1 of 7 areas
Following instructions
Claude Sonnet 4.6 is the stronger all-rounder.Nova 2.0 Pro Preview is cheaper and better at following instructions.
Scores updated · 18 tests both models report · How we compare
Where each one wins
Tests won in each of the seven 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 software30Claude Sonnet 4.63 of 3 tests
- ReasoningHard problems that need careful thinking30Claude Sonnet 4.63 of 3 tests
- AgentsCarrying out multi-step tasks on its own20Claude Sonnet 4.62 of 2 tests
- FactsGetting facts right instead of making them up20Claude Sonnet 4.62 of 2 tests
- Images and chartsUnderstanding pictures, charts and video10Claude Sonnet 4.61 of 1 test
- Long documentsFinding answers in very long texts10Claude Sonnet 4.61 of 1 test
- Following instructionsDoing exactly what it is asked01Nova 2.0 Pro Preview1 of 1 test
Images and charts, 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.
Nova 2.0 Pro Preview costs 38% 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 4.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+42.6points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+41.6points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+28.8points ahead
Where Nova 2.0 Pro Preview pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+22.4points ahead
Every test, side by side
All 18 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4.6
- Terminal-Bench 2.1Claude Sonnet 4.6 by 41.671.229.6+41.6
- Terminal-Bench HardClaude Sonnet 4.6 by 28.85324.2+28.8
- SciCodeClaude Sonnet 4.6 by 7.450.142.7+7.4
AgentsClaude Sonnet 4.6
- GDPValClaude Sonnet 4.6 by 36.736.70+36.7
- τ-Bench V3 · BankingClaude Sonnet 4.6 by 26.834.47.6+26.8
ReasoningClaude Sonnet 4.6
- Humanity's Last ExamClaude Sonnet 4.6 by 24.233.69.4+24.2
- GPQA DiamondClaude Sonnet 4.6 by 987.578.5+9
- CritPtClaude Sonnet 4.6 by 3.13.10+3.1
FactsClaude Sonnet 4.6
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.6 by 42.651.69+42.6
- AA-Omniscience · AccuracyClaude Sonnet 4.6 by 18.940.922+18.9
Images and chartsClaude Sonnet 4.6
- MMMU-ProClaude Sonnet 4.6 by 8.873.364.5+8.8
Following instructionsNova 2.0 Pro Preview
- IFBenchNova 2.0 Pro Preview by 22.456.679+22.4
Other results5 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 4.6 by 61.212.2-49+61.2
- Artificial Analysis Coding IndexClaude Sonnet 4.6 by 296334+29
- AA Agentic IndexClaude Sonnet 4.6 by 26.133.17+26.1
- τ²-Bench Telecom (AA run)Nova 2.0 Pro Preview by 1775.792.7+17
- AA IntelligenceClaude Sonnet 4.6 by 15.930.114.2+15.9
Questions people ask
Which is better, Claude Sonnet 4.6 or Nova 2.0 Pro Preview?
Claude Sonnet 4.6 wins six of the seven areas where both have results: coding, agents, reasoning, facts, images and charts and long documents. Nova 2.0 Pro Preview wins following instructions, and costs 38% less.
Which is better for coding?
Claude Sonnet 4.6. It wins 3 of the 3 coding tests both models report; Nova 2.0 Pro Preview wins none.
Which is cheaper?
Claude Sonnet 4.6 costs $3.00 per million input tokens and $15.00 per million output tokens; Nova 2.0 Pro Preview costs $1.25 and $10.00. That makes Nova 2.0 Pro Preview about 38% cheaper for the same work.
How do you compare the two?
We use the 18 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 5 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.