新版舊版
17 tools, three companies, one page
The latest from Anthropic / OpenAI / Google (updated 2026/10/7) plus a full free vs paid comparison of 17 tools. You can use them with no coding.
黃志弘 Leo
Published 2026-04-27 · Updated 2026-10-07 · 1 readers
Through 2026 the three companies took turns bombing us with news: new models, one pulled and brought back, tiers split by size, and another round of flagship swaps in September. Feeling buried when you open your phone is normal. This page lines up 17 tools for you in one go (updated 2026/10/7).
This page covers the 17 tools that are usable right now. Every entry says whether there is a free tier and what the cheapest paid plan costs, so you can read it in 5 minutes and decide. Models and prices from Anthropic, OpenAI and Google, plus prices for Midjourney, Kling and Perplexity, were checked against their official pages on October 7, 2026. I could not read current prices for Seedream or Seedance from their official sites this time, so check their sites before you pay.
J U M P T O W H A T I N E E D
What's new from the three companies, June through September
ANTHROPIC
Claude
Fable 5.1 (flagship)
- 6/9Fable 5 released
- 6/12Delisted by the US government
- 6/30Sonnet 5
- 7/1Fable 5 back worldwide
- 7/24Opus 5 takes over as the workhorse
- 9/1Fable 5.1 / Mythos 5.1 take the flagship spot
Best writer
Deepest thinker
Deepest thinker
OPENAI
ChatGPT
GPT-6 Astra (flagship)
- 7/9GPT-5.6 public release
- 8/6Luna free tier, unlimited text
- 9/3GPT-6 Astra takes the flagship spot
- Images 2.0 keeps getting updates
- 4/26Sora shut down (API closed 9/24)
- Flex, cheap batch processing
Does everything
Widest ecosystem
Widest ecosystem
GOOGLE
Gemini
3.1 Pro (flagship)
- July3.5 Pro delayed a third time, still unreleased
- 7/213.6 Flash lands as the stand-in
- 8/133.7 Flash out
- 9/23.8 Flash out (the strongest workhorse)
- 9/153.8 Live / Live Extended Thinking
- Veo 3.1 (Lite included)
- 8/17Imagen 4 API shut down
- Spark, the on-phone model
- NotebookLM keeps expanding
Freshest info
Best video
Best video
Which one for which job?
| What you want to do | Use this |
|---|---|
| Long writing, editing drafts | Claude |
| Coding | Claude Code |
| Chatting, hunting for ideas | ChatGPT |
| Images (Chinese text / detail) | ChatGPT Images |
| Images (4K + text) | Nano Banana |
| Video (big international production) | Google Flow |
| Video (Chinese-language scenes) | Seedance 2.0 |
| Finding the newest information | Gemini |
| Study notes from a PDF | NotebookLM |
| Fact checks with citations | Perplexity |
All 17 tools in detail
Writing / chat / reasoning
3 tools ・ every one has a usable free tier
Claude
by Anthropic ・ flagship Fable 5.1 (out 9/1) plus Opus 5.5 (out 9/22) and Sonnet 5.5 (out 9/28)
Free tier
Strengths: Long-form logic, code quality, 128K output
Price: Pro $20/mo, Max $100-200/mo
Who it's for: People who write long pieces, edit drafts, code, or do research
M Y T A K EWhen I write an IG caption I hand it to Claude for the second pass. The logic is tighter than ChatGPT. The free tier has a daily message cap, but for normal use it is enough.
Rewrite this text so it sounds spoken and conversational, under 200 words.
Original: [paste your draft]ChatGPT
by OpenAI ・ flagship GPT-6 Astra (out 9/3); Work and Codex also offer GPT-6.1 Sol (out 9/29) and GPT-6 Luna
Free tier
Strengths: Idea generation, image generation, voice chat, and the widest plugin/GPTs ecosystem
Price: Go $8/mo, Plus $20/mo, Pro $100 / $200 / $500/mo
Who it's for: Anyone who wants one tool that does a bit of everything, and people who build custom GPTs
M Y T A K EFor a new carousel topic I always ask ChatGPT for 10 directions first. It gets creative faster than Claude, but Claude still writes the smoother draft.
I have a day job and I want to start an IG side project. Give me 10 post topics that fit me, each with a one-line hook.Gemini
by Google ・ flagship Gemini 3.1 Pro (3.5 Pro delayed, still unreleased) plus 3.8 Flash (out 9/2)
Free tier
Strengths: Finding the newest information (wired into Google Search), long-context and multimodal reasoning, and video generation (Google Flow)
Price: AI Plus $4.99, Pro $19.99, Ultra $99.99 or $199.99/mo (US prices); in Taiwan Plus is NT$165, Pro NT$650, Ultra from NT$3,300
Who it's for: People checking recent events, doing research, or wanting Google Workspace integration
M Y T A K EFor current events I use Gemini. It has one extra layer that ChatGPT lacks, Google Search. Since the Mac app landed on 4/15 I can call it with Option+Space, faster than opening a browser.
Summarize the latest AI policy moves in Taiwan as of July 2026, with Google source links.Coding
2 tools ・ Claude Code needs Pro, Codex runs on the free tier too (lower quota)
Claude Code
by Anthropic ・ agentic coding system
Needs Pro/Max
Strengths: Reads a whole codebase, refactors across files, runs tests on its own, 1M token context
Price: Claude Pro $20/mo, or Max $100/$200
Who it's for: Engineers shipping products, vibe coders, anyone building a side-project app with AI
M Y T A K EEvery product I have is written with Claude Code, including this site, leoaido.com. What it does better than Cursor or Copilot is read the whole project structure, so refactors do not break things.
Take a look at the overall architecture of this codebase, find 3 places worth refactoring and say why.Codex
by OpenAI ・ the free tier uses GPT-6 Luna; from Plus you can use GPT-6.1 Sol, the model OpenAI recommends for coding
Free tier
Strengths: OpenAI's own coding tool
Price: Works on the free tier (GPT-6 Luna only, lower quota). Plus at $20/mo adds GPT-6.1 Sol and cloud features such as automatic code review. You can also connect your own API key and pay per use
Who it's for: People who want a coding agent right inside ChatGPT, and the free quota is enough to try it
M Y T A K EBetween Codex and Claude Code I pick Claude Code, for output quality and for holding a big project in its head. Codex is good for pulling up mid-conversation in ChatGPT to fix one section.
[paste your function]
Why is this code slow? Give me concrete suggestions to speed it up and a rewritten version.Images
4 tools ・ 2 with free tiers (Nano Banana / Seedream) + 2 best paid for (ChatGPT Images / Midjourney)
ChatGPT Images
by OpenAI ・ version 2.0 hit the top of Image Arena when it came out on 4/21, and 2.5 shipped on 9/8
Plus recommended
Strengths: Best Chinese text rendering, best detail, hit #1 on Image Arena within 12 hours
Price: Built into ChatGPT. The free tier can make images in ChatGPT chats too, with a lower cap. If you use it a lot, or want Images with thinking, get Plus ($20/mo). In Codex, image generation is not on the free plan
Who it's for: People making Chinese posters, IG post images, or anything that needs readable Chinese text inside the picture
M Y T A K EI tried it the day it came out on 4/21. Chinese IG covers came back with clean type and no broken characters. Compared with DALL-E 3 it is a generation ahead, the upgrade most worth paying for that month.
Make an IG carousel cover on a beige background, gold serif title “AI Tool Map”, small subtitle “Updated 2026/9”, 1080×1350 portraitNano Banana
by Google ・ current versions are Nano Banana 2.1, 2 and Pro
Free tier
Strengths: 4K output, strong text rendering, fastest edit loop with Speed Mode
Price: Free tier inside the Gemini app; the Pro version comes with AI Pro at $19.99/mo
Who it's for: People who need to try several concepts fast, or want 4K resolution
M Y T A K ESpeed Mode really is fast, 4 versions in under 30 seconds to pick from. Chinese detail is still steadier on ChatGPT Images, so it depends which fits your workflow.
Generate a 4K minimalist editorial poster: warm beige gradient background, gold accent line under bold Chinese title Midjourney v8
by Midjourney ・ v8.2 is the default version from 2026/7
No free tier
Strengths: Best aesthetics, biggest community, mature Style Reference system
Price: Basic $10/mo (200 minutes of Fast GPU time a month), Standard $30/mo (unlimited Relax for images), Pro $60/mo; no free trial on the website or Discord, only a limited trial in the niji·journey app
Who it's for: People making art-style images, brand visuals, or chasing the best single-image quality
M Y T A K EStill the aesthetics king in 2026, but Chinese text is basically unusable. I never pick Midjourney for IG. I use it for avatars and brand key visuals.
minimalist editorial poster, warm beige paper texture, gold horizontal accent, bold serif Chinese typography vibe, magazine cover style --ar 4:5 --v 8Seedream 5.0
by ByteDance ・ the CapCut team
Free tier
Strengths: Generates with live web retrieval, 4K output, accurate Chinese posters
Price: The free tier has a daily quota; the Pro version runs on prepaid credits
Who it's for: People making Chinese social covers, marketing posters, or UI design that needs a Chinese context
M Y T A K EChinese scenes land better than Midjourney or Imagen (street food stalls, night markets, the metro), though the styling has less design polish than the international tools. First pick for local Taiwanese content.
Make an IG cover in Taiwanese night market style, theme “the small joys after work”, beige background with a faint neon glow, gold serif titleVideo
4 tools ・ Kling and Seedance have free tiers ・ Google Flow makes video from AI Plus ($4.99/mo) up
Google Flow (Veo 3.1 / Gemini Omni Flash)
by Google DeepMind ・ Flow now uses Gemini Omni Flash for video by default, and Veo 3.1 is still there
From AI Plus
Strengths: Video and sound come out together. Omni Flash takes text, images and video as references and lets you keep editing after it generates; Veo 3.1 makes 8-second clips up to 4K
Price: On the free plan, Flow only gives limited Nano Banana Pro image generation. Video starts at Google AI Plus: Plus $4.99/mo includes 200 Flow credits, Pro $19.99 includes 1,000, Ultra includes 10,000 or 25,000 (US prices; in Taiwan Plus is NT$165 and Pro NT$650)
Who it's for: People making brand ads, product promos, or anything that needs native audio and lip sync
M Y T A K EThe native audio really is ahead. Other tools are still doing lip sync in post, while Veo outputs an 8 second clip with sound straight away. Whether 8 seconds is enough depends on what you make.
Generate an 8-second cinematic shot: a Taiwanese coffee shop morning, warm sunlight through window, hands holding a phone, ambient cafe sounds. Vertical 9:16 portrait.Kling 3.0
by Kuaishou, China ・ out 2026/2/4
Free tier
Strengths: API from about $0.084 per second (3.0, 720p, no audio), multi-shot storyboards, up to 15 seconds per clip (shots can be chained)
Price: Has a free tier (the official site lists no fixed daily credits; free credits come with promotions); Standard monthly is $6.99 the first month, then $8.8/mo, with 660 credits a month
Who it's for: People pumping out short social video on a tight budget who need longer clips
M Y T A K EFor the cost of 1 Sora clip, Kling makes 3. For Chinese subtitles and lip sync in Taiwan it is weaker than Veo, but for pure visuals nothing beats the value.
Make a 15-second narrative storyboard: 3 shots of a person opening laptop / drinking coffee / smiling at screen, cinematic warm tones, Asian male in his 30sRunway Gen-4.5
by Runway
Free trial + paid
Strengths: Most flexible for stylized work, mature creator ecosystem, Director Mode
Price: Standard $15/mo, Pro $35/mo
Who it's for: Video creators, people making music videos, anyone experimenting with stylized looks
M Y T A K ERunway is the veteran of AI video. For hero content it looks better than Kling but costs more than Veo. If you are already a creator and budget is not the issue, Runway is still the first pick.
Director Mode: stylized POV walking through a Taipei night market at dusk, neon signs reflecting on wet street, slow motion, vaporwave aestheticSeedance 2.0
by ByteDance ・ out 2026/2
Free tier
Strengths: Omni-Reference solves face swapping for characters, 15 second multi-shot clips, native audio
Price: The free tier has a daily quota; Pro runs on prepaid credits
Who it's for: People making Chinese-language short video, needing character consistency, or shooting ad scripts
M Y T A K EOmni-Reference genuinely fixes the face-swap problem, the biggest pain point in AI video so far. For Chinese scenes and Asian faces no tool is more accurate than Seedance right now.
Use Omni-Reference: upload a portrait photo of me + a clip of Taipei street scenes + a sample of the camera style I want, then generate a 15-second video of “me walking through that street scene”Learning / research / knowledge work
4 tools ・ NotebookLM, Perplexity and Deep Research have free tiers ・ the former Cowork features need Pro/Max
Gemini Notebook (formerly NotebookLM)
by Google
Free tier
Strengths: Upload a PDF, video or web page and it becomes a study assistant, with an automatic Audio Overview and mind map
Price: Free tier gets 100 notebooks; AI Plus gets 200 and AI Pro 500, with usage limits that reset every 5 hours
Who it's for: Students, researchers, anyone processing meeting recordings or long articles
M Y T A K EI fed Mosseri's 6 official posts on the IG algorithm into NotebookLM and the summary came out 5 times faster than translating and organizing by hand. Audio Overview is great on the commute.
(upload a PDF or audio file first) Make me a 1-page cheat sheet listing the 5 most important points in the material + how to actually apply each onePerplexity
by Perplexity AI
Free tier
Strengths: Checks facts and attaches sources automatically, and picks a suitable model for each question
Price: Pro about US$17/mo billed yearly (check the site for the monthly price); switch between GPT, Claude, Gemini, Grok and other models
Who it's for: Journalists, researchers, anyone who needs a quick fact-check or is writing a paper
M Y T A K EWhen a post needs a number I start with Perplexity, then click into the source to confirm. Faster than Google Search, safer than ChatGPT because it does not make things up.
How many monthly active ChatGPT users did Taiwan have in 2026? Confirm it with at least 3 independent sourcesClaude (formerly Claude Cowork)
by Anthropic
Needs Pro/Max
Strengths: Scheduled tasks (recurring and on-demand), Mac and Windows desktop apps, in-line Charts added on 4/22
Price: From Claude Pro at $20/mo
Who it's for: Knowledge workers who want Claude running daily tasks on its own
M Y T A K EI have Cowork scheduled to scan Threads trends at 8 every morning, and waking up to the conclusions feels great. This is the early shape of a long-running agent.
Schedule: every morning at 8:00, scan the top 10 high-engagement posts under the Threads tags “AI side income” and “one-person company”, then send me a bulleted summaryDeep Research
by Google ・ built into Gemini, the free plan includes it
Free tier
Strengths: Hours of automated research, pulls many sources together, outputs a long document
Price: Included in the free Gemini plan, and in Google AI Pro ($19.99/mo)
Who it's for: Consultants, people doing industry analysis, writing research reports, or running investment due diligence
M Y T A K EI gave it a topic like "the size of the nail salon market in Taiwan in 2026" and 2 hours later got a full report with 30 source links. Much faster than a week of googling.
Do deep research on Taiwan's AI education market in 2026: market size, main players, target user segments, growth rate, with at least 20 source linksFree vs paid at a glance
Has a real free tier
- Claude (daily cap)
- ChatGPT (limited features)
- Gemini (cheapest free tier)
- Deep Research (in the free Gemini plan)
- Perplexity (basic search is free)
- Gemini Notebook (formerly NotebookLM, 100 notebooks)
- Nano Banana (inside the Gemini app)
- Kling 3.0 (no fixed daily credits listed; free credits come with promotions)
- Seedream 5.0 / Seedance 2.0 (daily quota)
- Codex (free tier works, lower quota)
- ChatGPT Images (lower cap on the free tier; Plus $20 recommended for regular use)
Paid only
- Midjourney v8 (from $10/mo, nothing free)
- Video in Google Flow (AI Plus from $4.99/mo)
- Claude Code (needs Pro $20)
- Claude scheduled tasks (formerly Cowork, needs Pro $20)
- Runway (the free plan gives 125 one-time credits; Standard from $15/mo)
Common questions
I know nothing yet. Which AI tool do I start with?
ChatGPT or Claude is the safest start. Both have a free tier, both look like a chat window, and you do not have to learn any commands. Writing copy, organizing notes, hunting for ideas, they all work. Use the free tier every day for a week, find the situation you reach for most, then decide whether to pay.
Are the free tiers good enough?
Depends what you do. For writing, chatting and research, the free tier covers 90% of it (Claude, ChatGPT, Gemini, Perplexity, NotebookLM and Kling all have usable free tiers). But images (Midjourney has nothing free), video in Google Flow (from AI Plus, $4.99/mo) or regular use of ChatGPT Images (Plus recommended) means paying.
Which AI is best at Chinese?
Writing Chinese: Claude > ChatGPT > Gemini, and the gap is small. Chinese images: ChatGPT Images (version 2.0 out 4/21, updated to 2.5 on 9/8, the strongest Chinese text rendering) plus Seedream 5.0 (accurate Chinese posters). Chinese video: Seedance 2.0, which leads the field on Asian faces and Chinese-language scenes.
Why were there so many AI updates in September 2026?
The first two weeks of September 2026 moved all three. Anthropic shipped Fable 5.1 and Mythos 5.1 on 9/1 (price held at $10/$50 per MTok, cache reads cut from $1.00 to $0.25). Anthropic followed with Opus 5.5 on 9/22 and Sonnet 5.5 on 9/28. OpenAI shipped GPT-6 Astra on 9/3, taking over from GPT-5.6 Sol as the flagship, then GPT-6 Sol and GPT-6 Luna on 9/22 and GPT-6.1 Sol on 9/29. Google shipped 3.8 Flash on 9/2 and 3.8 Live plus 3.8 Live Extended Thinking on 9/15. Gemini 3.5 Pro is still unreleased, so the Pro-tier flagship is still 3.1 Pro. Overall the rhythm holds at 1 big update every two weeks.
I only get 2 hours after work. Which one is worth learning?
Suggested order: (1) ChatGPT free tier for chat and ideas, (2) Claude for editing drafts and long writing, (3) NotebookLM to turn a PDF into study notes, (4) ChatGPT Images for IG post images. These 4 cover 80% of what a working person needs for content and learning, and the first 3 have free tiers.
Does Midjourney really have no free tier?
Right, Midjourney has no free tier, and no free trial on the website or Discord either; only the niji·journey mobile app has a limited trial. The cheapest Basic Plan is $10/mo (200 minutes of Fast GPU time a month), Standard is $30/mo (unlimited Relax for images), Pro is $60/mo. For free image generation, start with Nano Banana in the Gemini app, or Seedream 5.0 with its free daily quota. If you will use ChatGPT Images a lot, Plus ($20/mo) is the way to go.
Sora shut down. Can I still use it?
Sora's web and app versions shut down on 2026/4/26, and the API closed on 2026/9/24 too. OpenAI offers no replacement video API. For AI video go to Google Flow (Veo 3.1 / Gemini Omni Flash), Kling 3.0 (best value), Seedance 2.0 (Chinese scenes) or Runway Gen-4.5 (stylized).