Anthropic
Alternatives

Best cheaper alternatives to Claude 3 Sonnet

Looking for a cheaper or smarter replacement for Claude 3 Sonnet by Anthropic? We compared every LLM model with verifiable pricing on real input/output cost, context window, intelligence and modalities. 18 options are meaningfully cheaper.

Category: LLM Intelligence 45 200K context
Google
Gemini 2.5 Flash86% CHEAPER
Google
Better valueSmarter (68 vs 45)Larger contextMore modalitiesSame category
Google
Gemini 2.0 Flash97% CHEAPER
Google
Better valueSmarter (56 vs 45)Larger contextMore modalitiesSame category
OpenAI
o395% CHEAPER
OpenAI
Better valueSmarter (78 vs 45)Same category
DeepSeek
DeepSeek-R1 (Reasoner)96% CHEAPER
DeepSeek
Better valueSmarter (73 vs 45)Same category
Google
Gemini 1.5 Flash98% CHEAPER
Google
Better valueSmarter (48 vs 45)Larger contextMore modalitiesSame category
OpenAI
o195% CHEAPER
OpenAI
Better valueSmarter (71 vs 45)Same category
OpenAI
o4 Mini95% CHEAPER
OpenAI
Better valueSmarter (70 vs 45)Same category
Google
Gemini 1.5 Flash-8B99% CHEAPER
Google
Better valueLarger contextMore modalitiesSame category
OpenAI
GPT-4o95% CHEAPER
OpenAI
Better valueSmarter (60 vs 45)More modalitiesSame category
OpenAI
GPT 4.1 Mini95% CHEAPER
OpenAI
Better valueSmarter (53 vs 45)Larger contextSame category
DeepSeek
DeepSeek R184% CHEAPER
DeepSeek
Better valueSmarter (73 vs 45)Same category
DeepSeek
DeepSeek V392% CHEAPER
DeepSeek
Better valueSmarter (63 vs 45)Same category
DeepSeek
DeepSeek-V3 (Chat)98% CHEAPER
DeepSeek
Better valueSmarter (63 vs 45)Same category
OpenAI
GPT 4.1 Nano95% CHEAPER
OpenAI
Better valueLarger contextSame category
Alibaba Cloud
Qwen Plus90% CHEAPER
Alibaba Cloud
Better valueSmarter (55 vs 45)Same category
Meta
Llama 3.2 90B Vision96% CHEAPER
Meta
Better valueSmarter (54 vs 45)Same category
Together AI
Llama 3.3 70B Instruct86% CHEAPER
Together AI
Better valueSmarter (58 vs 45)Same category
Alibaba Cloud
Qwen2.5 Math 72B Instruct86% CHEAPER
Alibaba Cloud
Better valueSmarter (58 vs 45)Same category

Costs are normalised to a blended 1M-token workload (or per-unit price for non-token models) using each model's latest verified pricing. Models with pricing under review are excluded from recommendations.