Best AI for research

Finding information, checking facts and pulling it all together.

Best overall, and best for $25 a month or lessToo close to call

GPT-6 AstraGet it with ChatGPT Plus, $20 a month (only in ChatGPT's Work mode and Codex, not normal chat)
GPT-5.6 SolGet it with ChatGPT Plus, $20 a month

1st and 2nd of 11 models for this task

Best freeToo close to call

Gemini 3.6 FlashGet it with Gemini Free
Kimi K3Get it with Kimi Free

3rd and 4th of 11 models for this task

Keep in mind Facts and expert-question tests help, but nothing yet tests research reports or citations.

Based on tests of related skills. The picks follow a fixed rule using the charts below and each app's plans.

Already paying for one? See what each plan gives you here

App and plan	Price	Best model on it for this	Position
ChatGPT Free	free	GPT-5.6 Luna	9th of 11
ChatGPT Go	$8 a month	GPT-5.6 Luna	9th of 11
ChatGPT Plus	$20 a month	GPT-6 Astra Only in ChatGPT's Work mode and Codex, not normal chat	1st of 11
ChatGPT Pro	$100 a month	GPT-6 Astra Called GPT-6 Pro in the app	1st of 11
Claude Free	free	Claude Sonnet 5	10th of 11
Claude Pro	$20 a month	Claude Sonnet 5 Its main model, Claude Opus 5.5, is too new to score yet.	10th of 11
Claude Max	$100 a month	Claude Fable 5.1 For up to half your weekly allowance Its main model, Claude Opus 5.5, is too new to score yet.	5th of 11
Gemini Free	free	Gemini 3.6 Flash	3rd of 11
Google AI Plus	$4.99 a month	Gemini 3.6 Flash	3rd of 11
Google AI Pro	$19.99 a month	Gemini 3.8 Flash	6th of 11
Google AI Ultra	$99.99 a month	Gemini 3.8 Flash	6th of 11
Kimi Free	free	Kimi K3	4th of 11
Kimi Moderato	$19 a month	Kimi K3	4th of 11

Norm's guide

What I'd tell a friend who asked.

What AI is good at

Summarising long articles and reports.
Explaining a topic you're new to.
Comparing a few options side by side.

Where it trips up

Inventing sources, quotes or statistics.
Out-of-date information.
Presenting a guess as if it were a fact.

How to get a good result

Ask for sources, then open them yourself.
Use a model that can search the web for anything recent.
Double-check any number you plan to repeat.

How the models compare

Every model we could score for research, combining the tests below.

Overall for research

Position on a combined score from 3 tests. The longer the bar, the further ahead · Higher is better

On the tests we have, GPT-6 Astra comes out on top, followed by GPT-5.6 Sol and Gemini 3.6 Flash.

GPT-6 Astra · OpenAI1st

GPT-5.6 Sol · OpenAI2nd

Gemini 3.6 Flash · Google DeepMind3rd

Kimi K3 · Moonshot AI4th

Claude Fable 5.1 · Anthropic5th

Gemini 3.8 Flash · Google DeepMind6th

Gemini 3.1 Pro · Google DeepMind7th

DeepSeek V4 Pro · DeepSeek8th

GPT-5.6 Luna · OpenAI9th

Claude Sonnet 5 · Anthropic10th

Claude Haiku 4.5 · Anthropic11th

The evidence

Answering short factual questions correctly

Tests a related skill · Short questions that each have one checkable answer. · Higher is better

GPT-6 Astra and Gemini 3.1 Pro are neck and neck at the top, followed by Claude Fable 5.1 and Gemini 3.8 Flash; GPT-6 Astra costs about 4 times as much.

GPT-6 Astra · OpenAI76%

Gemini 3.1 Pro · Google DeepMind74%

Claude Fable 5.1 · Anthropic71%

Gemini 3.8 Flash · Google DeepMind70%

GPT-5.6 Sol · OpenAI70%

Gemini 3.6 Flash · Google DeepMind66%

DeepSeek V4 Pro · DeepSeek53%

Kimi K3 · Moonshot AI51%

GPT-5.6 Luna · OpenAI41%

Claude Sonnet 5 · Anthropic34%

Claude Haiku 4.5 · Anthropic13%

Source: Google DeepMind via Epoch AI · CC BY 4.0

Humanity's Last Exam

Tests a related skill · Very hard questions written by experts across many subjects. · Higher is better

GPT-6 Astra leads, followed by Claude Fable 5.1 and Gemini 3.1 Pro.

GPT-6 Astra · OpenAI55%

Claude Fable 5.1 · Anthropic47%

Gemini 3.1 Pro · Google DeepMind46%

Gemini 3.8 Flash · Google DeepMind45%

Source: Center for AI Safety, Scale AI via Epoch AI · CC BY 4.0

People's votes on expert questions

Tests a related skill · People compared two anonymous answers to questions needing expert knowledge. · Higher is better

Claude Opus 5.5 and Gemini 3.8 Flash are neck and neck at the top, followed by Claude Fable 5.1 and Kimi K3; Claude Opus 5.5 costs about 5 times as much.

Claude Opus 5.5 · Anthropic1st

Gemini 3.8 Flash · Google DeepMind2nd

Claude Fable 5.1 · Anthropic3rd

Kimi K3 · Moonshot AI4th

GPT-5.6 Sol · OpenAI5th

DeepSeek V4.1 Flash · DeepSeek6th

Claude Sonnet 5 · Anthropic7th

Gemini 3.6 Flash · Google DeepMind8th

GPT-6 Astra · OpenAI9th

Gemini 3.1 Pro · Google DeepMind10th

GPT-5.6 Luna · OpenAI11th

DeepSeek V4 Pro · DeepSeek12th

Claude Haiku 4.5 · Anthropic13th

Gemini 3.5 Flash-Lite · Google DeepMind14th

Source: LMArena · votes as of 25 Sept 2026

If you build with it

What the makers charge developers who use the models directly. Using an app? The plans above are what you pay.

Cost per 1,000 typical requests

List price, about 1,500 words in and 500 out per request · Lower is better

GPT-6 Luna is cheapest, followed by DeepSeek V4.1 Flash and GPT-5.6 Luna.

GPT-6 Luna · OpenAI$0.55

DeepSeek V4.1 Flash · DeepSeek$0.72

GPT-5.6 Luna · OpenAI$1.24

DeepSeek V4 Pro · DeepSeek$1.48

Gemini 3.5 Flash-Lite · Google DeepMind$2.35

Gemini 3.8 Flash · Google DeepMind$4.13

Gemini 3.6 Flash · Google DeepMind$4.13

Claude Haiku 4.5 · Anthropic$5.50

Claude Sonnet 5 · Anthropic$11

GPT-6 Sol · OpenAI$11

Gemini 3.1 Pro · Google DeepMind$12

Kimi K3 · Moonshot AI$17

GPT-5.6 Sol · OpenAI$22

Claude Opus 5.5 · Anthropic$22

GPT-6 Astra · OpenAI$55

Claude Fable 5.1 · Anthropic$55

Source: models.dev · MIT