The best value AI models for each kind of job, ranked by ability per cent. Every model listed scores at least 60/100 at the task, so you're not trading away quality to save money. Updated 2026-10-11.
Cheapest good AI for reasoning
Ling 3.1 Flash: reasoning score 88, 0.021¢ a chat (4,846 per NZ$1)
Ling 3.0 Flash VL: reasoning score 75, under 0.01¢ a chat (28,300 per NZ$1)
Qwen 3.8 Max: writing score 84, 0.36¢ a chat (275 per NZ$1)
Claude Opus 5.5: writing score 97, 2.22¢ a chat (45 per NZ$1)
Cheapest good AI for following instructions
GLM 5.3 Flash: following instructions score 78, 0.040¢ a chat (2,505 per NZ$1)
DeepSeek V4.1 Flash: following instructions score 82, 0.099¢ a chat (1,009 per NZ$1)
MiMo V2.5: following instructions score 64, 0.018¢ a chat (5,549 per NZ$1)
Cheapest good AI for agentic tasks
Ling 3.0 Flash VL: agentic tasks score 75, under 0.01¢ a chat (28,300 per NZ$1)
Qwen 3.8 Max: agentic tasks score 97, 0.36¢ a chat (275 per NZ$1)
GLM 5.3 Flash: agentic tasks score 77, 0.040¢ a chat (2,505 per NZ$1)
FAQ
What is the cheapest AI model right now?
Ling 3.0 Flash VL at under 0.01¢ per typical chat, then Laguna S 2.1 (under 0.01¢) and GPT-6 Luna (0.013¢). Cheapest isn't always best value, so the lists above weigh ability against price.
How is 'cost per chat' measured?
We send the same set of realistic prompts to every model and record the tokens it actually uses, including hidden reasoning tokens, then apply its list price. It's what a normal chat really costs, in NZ cents.