Model	SWE-bench Verified % resolved	Terminal-Bench 2.1 task success	DeepSWE % resolved	GPQA Diamond accuracy	LiveCodeBench Pro Codeforces Elo	Humanity's Last Exam accuracy	LiveBench score	SWE-bench Pro % resolved	Berkeley Function Calling Leaderboard accuracy	MMMU-Pro accuracy	LMArena source-defined rating	ARC-AGI-3 accuracy	ARC-AGI-2 accuracy	FrontierMath accuracy	AIME (OTIS Mock) accuracy	SimpleQA Verified accuracy	GBA Eval overall score	WeirdML average accuracy	SWE-fficiency speedup score	KernelBench (Hard) fast₁	MCP Atlas pass rate	Artificial Analysis Intelligence Index Index	Epoch Capabilities Index Index	Aider Polyglot % correct	SWE-rebench Resolved rate (pass@1)	MMLU-Pro accuracy	OSWorld task success rate	GAIA: A Benchmark for General AI Assistants accuracy	BrowseComp accuracy	τ²-bench (Telecom) pass^1	AIME 2026 accuracy	MathVista accuracy	Video-MME accuracy	GDPval Elo	LiveCodeBench Pass@1	METR Task-Completion Time Horizons 50% time horizon	SWE-Lancer IC SWE pass rate (Diamond)	SciCode accuracy	MMMU (Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark) accuracy	BIG-Bench Extra Hard harmonic mean accuracy	AA-Omniscience: Knowledge and Hallucination Benchmark AA-Omniscience Index	IFBench accuracy	MultiChallenge accuracy	RULER accuracy	OpenAI-MRCR v2 (Multi-Round Coreference Resolution) accuracy (mean SequenceMatcher similarity)	LongBench v2 accuracy	Global-MMLU accuracy	MMLU-ProX accuracy	Video-MMMU accuracy	WebDev Arena Elo	Search Arena Elo	Arena-Hard-Auto v2.0 % win rate	EQ-Bench Creative Writing v3 Elo	Design Arena Elo	AgentHarm Harm score	AgentDojo utility under attack	AILuminate AI Safety Benchmark Safety grade	MASK (Model Alignment between Statements and Knowledge) Honesty score	MCP-Universe Overall Success Rate	CharXiv accuracy	OCRBench v2 accuracy	ScreenSpot-Pro accuracy	FACTS Grounding Grounding accuracy	BigCodeBench calibrated Pass@1	SWE-bench Multilingual % resolved	SWE-bench Multimodal % resolved	SuperGPQA accuracy	EnigmaEval accuracy	ZeroBench accuracy	IMO-Bench accuracy	PutnamBench Problems solved	MathArena HMMT February 2026 accuracy	FrontierMath Tier 4 accuracy	Vectara Hallucination Leaderboard Hallucination Rate	Gray Swan Arena (Agent Red-Teaming / Indirect Prompt Injection) Attack Success Rate (ASR)	PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts Difficulty-Weighted Accuracy (DW-ACC)
Amazon Nova 2 Pro Amazon Nova (Nova 2)	70.0%	—	—	81.4%	—	—	—	—	61.6%	63.5%	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	92.7%	—	—	—	—	—	—	—	—	—	—	—	79.0%	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—
Amazon Nova 2 Lite Amazon Nova (Nova 2)	64.5%	—	—	79.6%	—	—	—	—	60.3%	61.8% (MMMU-Pro)	—	—	—	—	—	—	—	—	—	—	24.6%	—	—	—	—	—	—	—	—	—	—	—	—	—	34.6%	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	5.1%	—	—

Scores are not normalised across benchmarks. Each column uses its own metric. Compare columns independently.