Hayz Labs · https://hayzlabs.com/reviews/rankings/research/
reviews · best for a job · 15 scored
Best for research and long documents
Hard questions, search, long PDFs and citing correctly. Ranked by score; on a tie, the cheaper one goes first. 100 is the best available today in this job.
Picked one? Research with AI and keep the sources shows how to get good results from it.
New to this? What the scores and labels mean
- 0 to 100
- Our score for one job, like writing or coding, compared with everything else out today. 100 is the best available today. For AI models we build it from public data, such as Arena's blind votes and Epoch AI's independent tests; apps and tools are scored on our own checklist. The same tool can score differently for different jobs, and where two likely ranges overlap, we call it a tie.
- Our verdict
- Use it: a top choice for most people. Worth trying: good, and worth a go. Depends: good for some people or one kind of job. Wait: too early to tell, or not out for everyone yet. Skip: something else does the job better.
- First look
- Just released. We say what it is and whether to bother, and give it a score once public results exist.
- No score yet
- None of the public sources we use covers it yet, or the only evidence is from a source whose numbers we can't use, so we describe it in words instead.
- Best buy
- Nothing cheaper scores as well for that job.
- Best free
- The best score for that job that costs nothing, through a free plan or by running on your own computer.
- Pay as you go
- No monthly plan: you pay for each use, usually through a developer account.
- Checked
- The day we last confirmed the score and the price against the sources. Prices change often, so check the date.
- AI model
- The engine underneath, like GPT or Claude. You use it through an app or, if you code, directly.
- Open source
- Free to download and run on your own computer, so nothing leaves your machine.
The short answer
For hard questions and long documents, Claude Opus 5.5 (97), Gemini 4 Argon (97) and Claude Sonnet 5.5 (96) score highest of the models we review on Arena's Expert board, with ranges that overlap, so treat them as tied.
- Best
- Gemini 4 Argon95$4.99/motied with 7 others
- Best free
- MiMo-V2.6-Pro94Free
Read our full take
Argon is the cheapest way in, with Google AI Plus at $4.99/month, though its votes are still early. From other makers, MiMo-V2.6-Pro scores 94 at $0.13 per Artificial Analysis Intelligence Index task, with free MIT weights that are sized for a GPU cluster rather than a home computer, and GPT-6.1 Sol scores 91 at $0.72 per index task at max effort, against $5.46 for Sonnet 5.5 at max. The free Claude plan, scored on our checklist, gets 80 at $0. We would not pay for Claude Fable 5.1 (93) or GPT-6 Astra (89) for this job: both cost $10 / $50 per 1M input / output tokens and score no higher than cheaper models. To watch: Gemini Notebook (formerly NotebookLM) is free for answering from your own files but not scored yet.
Our opinion, checked 11 Oct. The picks come straight from the scores below. How we score
This take recommends a product made by Anthropic, whose Claude models help draft our reviews. It also names the best pick from another maker.
Everything we cover for this job, best score first. Scores whose ranges overlap are tied, so choose on price and fit.
- 1Gemini 4 ArgonGoogle
Google's top model, first in Arena's text votes, now in Google AI Plus from $4.99/month; votes are still early.
Best for Everyday chat and writing at the top of Arena votes, from $4.99 a monthYou give up Votes still early; ranks 10th for web app coding, and API price doubles on 1 Dec
9588–98$4.99/mo - 2Claude Opus 5.5Anthropic
Top public pick for coding, agents and hard questions at a mid price; skip it for bulk work where Sonnet 5.5 is enough.
Best for Coding and agent jobs; ranks first on the web app and agent boardsYou give up Twice Sonnet 5.5's token price; thinking can't be turned off
9488–99$20/mo - 3MiMo-V2.6-ProXiaomi
Top open-weights model on the Artificial Analysis index, at cents per task; closed leaders still rank well above it.
Best for Cheap coding and agent work, or an MIT model to host yourselfYou give up Self-hosting needs 8 to 16 GPUs; data on Xiaomi's platform is stored in China
9486–100Free - 4GPT-6.1 SolOpenAI
Near-top coding and agent work at a fifth of GPT-6 Astra's price; pick Claude Opus 5.5 or Gemini for creative writing.
Best for Coding and agent work near the top without top-model pricesYou give up Weak at creative writing (rank 26); not on ChatGPT Free or Go
9487–99$20/mo - 5GPT-6 AstraOpenAI
Second on WebDev and Agent boards, but GPT-6.1 Sol is close for a fifth of the price; middling at creative writing.
Best for The hardest coding and agent jobs, where a small lead is worth paying forYou give up Five times GPT-6.1 Sol's token price; middling at creative writing
9388–96$20/mo - 6Claude Sonnet 5.5Anthropic
Near-top coding, agents and research for half Opus 5.5's token price; for writing, pick Opus 5.5 or Gemini 3.8 Flash.
Best for Everyday coding and agents with near-top results on a budgetYou give up Behind the leaders on creative writing; breaking API changes from Sonnet 5
9083–94Free - 7Gemini 3.8 FlashGoogle
Low-cost Google model near the top for writing and maths; rankings are preliminary and API prices double in January.
Best for Writing, maths and video or audio input at a low token priceYou give up Weak at web app coding (rank 31); API prices double on 1 Jan 2027
8884–92$19.99/mo - 8Claude Fable 5.1Anthropic
Very strong, but Opus 5.5 ranks higher on most boards for 40% of the price; keep Fable for cases Opus misses.
Best for Teams whose hardest long tasks still fail on Opus 5.5You give up Slower than Opus 5.5, 2.5 times its token price, and needs 30-day data retention
8882–94$100/mo - 9ClaudeAnthropic
Free runs Sonnet 5.5, near the top for coding and hard questions. Pay for Pro for agents and for writing with Opus 5.5.
Best for Free help with coding and hard questions; Pro for agent work and writingYou give up Free plan's model is weaker at creative writing; usage limits are not given in numbers
80ChecklistFree - 10GLM-5.3-FlashZ.ai (Zhipu)
Very cheap MIT model, 13th on arena.ai Math; for coding and agents MiMo-V2.6-Pro ranks higher for less per task.
Best for Cheap maths-heavy or bulk work through an API or your own serversYou give up Mid-table for coding and weak at creative writing; hosting needs a server
7470–77Free - 11PerplexityPerplexity
Good for cited web answers on Pro with GPT-6.1 Sol; skip it if you want free help, where Claude scores higher.
Best for Quick web research where you want sources to check each answerYou give up Trains on chats by default; opting out does not remove data already collected
70Checklist$17/mo - 12Muse Spark 1.3Meta
Solid coding and agent model at a low token price; GPT-6.1 Sol and Claude rank higher if quality matters most.
Best for Low-cost coding and agent work through Meta's APIYou give up Ranks well below GPT-6.1 Sol on agent tasks; the cheapest tier uses your data
6861–75$5/mo - 13Mistral Large 4Mistral AI
Mistral's 1T preview scores no higher than 72 in any job; weights due 31 Oct. Use GLM-5.3 or Kimi K3 now.
6256–67Pay as you go - 14ChatGPT (GPT-6)OpenAI
Good for agent work and coding on Plus with GPT-6 Astra. For writing, Gemini AI Pro and Claude Pro score higher.
Best for People who want one app for agent tasks and coding on the $20 Plus planYou give up Writing scores 50
60Checklist$20/mo - 15Gemini appGoogle
Good for writing on AI Pro with 3.8 Flash; weak at coding and hard questions, where Claude and ChatGPT score higher.
Best for People who already pay Google for storage and want a writing helperYou give up Free plan runs a much weaker model; usage limits are not published in numbers
50Checklist$19.99/mo
Score against cost
Up is a better score, left is cheaper. Plans are the cheapest that unlock the scored feature. 1 without a monthly price is left off; see Pay per use.
Research: free gets you a score of 94. $5 a month gets you 95; paying more buys no better score.
Names in one box share a score band and a price step; each shows its own score, best first.
Best value: MiMo-V2.6-Pro (94, free) · Gemini 4 Argon (95, $4.99/mo)
Not on the map, as they have no monthly price: Mistral Large 4 (62).
Show as a table
| Name | Score | Cheapest plan | Price step | Best value |
|---|---|---|---|---|
| Gemini 4 Argon | 95 · likely 88–98 | $4.99 a month | Under $10 | Yes |
| MiMo-V2.6-Pro · open source | 94 · likely 86–100 | Free | Free | Yes |
| Claude Opus 5.5 | 94 · likely 88–99 | $20 a month | $10 to $29 | |
| GPT-6.1 Sol | 94 · likely 87–99 | $20 a month | $10 to $29 | |
| GPT-6 Astra | 93 · likely 88–96 | $20 a month | $10 to $29 | |
| Claude Sonnet 5.5 | 90 · likely 83–94 | Free | Free | |
| Gemini 3.8 Flash | 88 · likely 84–92 | $19.99 a month | $10 to $29 | |
| Claude Fable 5.1 | 88 · likely 82–94 | $100 a month | $100 or more | |
| Claude | 80 | Free | Free | |
| GLM-5.3-Flash · open source | 74 · likely 70–77 | Free | Free | |
| Perplexity | 70 | $17 a month | $10 to $29 | |
| Muse Spark 1.3 | 68 · likely 61–75 | $5 a month | Under $10 | |
| Mistral Large 4 · open source | 62 · likely 56–67 | No monthly price | ||
| ChatGPT (GPT-6) | 60 | $20 a month | $10 to $29 | |
| Gemini app | 50 | $19.99 a month | $10 to $29 |
Up is a better score, left is cheaper to run. Cost to run one typical agent task (arena.ai, middle of the range), for the 10 scored here that have one; the others are left off rather than mixed with figures from another test.
| # | Name | Score | Source | Cheapest plan | Our verdict | Why |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 ArgonGoogle · AI model | 95likely 88–98Tied with 7 others | Our score | $4.99/mo | Use itBest buy | Google's top model, first in Arena's text votes, now in Google AI Plus from $4.99/month; votes are still early. |
| 2 | Claude Opus 5.5Anthropic · AI model | 94likely 88–99Tied with 7 others | Our score | $20/mo | Use it | Top public pick for coding, agents and hard questions at a mid price; skip it for bulk work where Sonnet 5.5 is enough. |
| 3 | MiMo-V2.6-ProXiaomi · Open source | 94likely 86–100Tied with 7 others | Our score | Free | Worth tryingBest buyBest free | Top open-weights model on the Artificial Analysis index, at cents per task; closed leaders still rank well above it. |
| 4 | GPT-6.1 SolOpenAI · AI model | 94likely 87–99Tied with 7 others | Our score | $20/mo | Use it | Near-top coding and agent work at a fifth of GPT-6 Astra's price; pick Claude Opus 5.5 or Gemini for creative writing. |
| 5 | GPT-6 AstraOpenAI · AI model | 93likely 88–96Tied with 7 others | Our score | $20/mo | Depends | Second on WebDev and Agent boards, but GPT-6.1 Sol is close for a fifth of the price; middling at creative writing. |
| 6 | Claude Sonnet 5.5Anthropic · AI model | 90likely 83–94Tied with 7 others | Our score | Free | Use it | Near-top coding, agents and research for half Opus 5.5's token price; for writing, pick Opus 5.5 or Gemini 3.8 Flash. |
| 7 | Gemini 3.8 FlashGoogle · AI model | 88likely 84–92Tied with 7 others | Our score | $19.99/mo | Worth trying | Low-cost Google model near the top for writing and maths; rankings are preliminary and API prices double in January. |
| 8 | Claude Fable 5.1Anthropic · AI model | 88likely 82–94Tied with 7 others | Our score | $100/mo | Skip | Very strong, but Opus 5.5 ranks higher on most boards for 40% of the price; keep Fable for cases Opus misses. |
| 9 | ClaudeAnthropic · App | 80 | Our checklist | Free | Worth trying | Free runs Sonnet 5.5, near the top for coding and hard questions. Pay for Pro for agents and for writing with Opus 5.5. |
| 10 | GLM-5.3-FlashZ.ai (Zhipu) · Open source | 74likely 70–77Tied with #12 | Our score | Free | Worth trying | Very cheap MIT model, 13th on arena.ai Math; for coding and agents MiMo-V2.6-Pro ranks higher for less per task. |
| 11 | PerplexityPerplexity · App | 70 | Our checklist | $17/mo | Worth trying | Good for cited web answers on Pro with GPT-6.1 Sol; skip it if you want free help, where Claude scores higher. |
| 12 | Muse Spark 1.3Meta · AI model | 68likely 61–75Tied with #10, #13 | Our score | $5/mo | Worth trying | Solid coding and agent model at a low token price; GPT-6.1 Sol and Claude rank higher if quality matters most. |
| 13 | Mistral Large 4Mistral AI · AI model | 62likely 56–67Tied with #12 | Our score | Pay as you go | Wait | Mistral's 1T preview scores no higher than 72 in any job; weights due 31 Oct. Use GLM-5.3 or Kimi K3 now. |
| 14 | ChatGPT (GPT-6)OpenAI · App | 60 | Our checklist | $20/mo | Depends | Good for agent work and coding on Plus with GPT-6 Astra. For writing, Gemini AI Pro and Claude Pro score higher. |
| 15 | Gemini appGoogle · App | 50 | Our checklist | $19.99/mo | Depends | Good for writing on AI Pro with 3.8 Flash; weak at coding and hard questions, where Claude and ChatGPT score higher. |
Nothing matches all of that. , or try everything.
What changed
New leaders and rating changes in research; our working is under Details. All news
- Worse: Claude Fable 5.1: research rating moved down.
Details
New tie rule: models tied on the Artificial Analysis index take the lowest rank in the tie, so research is rated on that.