Ranking the Best AI Tools of 2026: What’s Actually Worth Using

There are more AI tools than ever in 2026—and most of them are fighting for the same $20–$200 a month in your budget. Some are genuinely game‑changing. Others are just productivity theater with a nice UI. This guide breaks down the major players by category and ranks them from S to D tier so you know what’s actually worth investing in.

Chatbots: Your Daily Driver AI

The first decision most people make is simple: which chatbot should I pay for every month? Here’s how the big four stack up as general-purpose assistants.

ChatGPT: Solid but Middle of the Road

Tier: B

ChatGPT is still where most people start. Its latest model (GPT‑5.4) is strong, and for $20/month you get a dependable, versatile chatbot plus high‑quality image generation with the new Image Gen model.

The downside: it has become the “average” chatbot. If your prompts are weak, the writing feels generic and instantly recognizable as ChatGPT. You’ll rarely hate the outputs, but you’ll rarely be blown away either.

Still, as a general tool for writing, brainstorming, and light coding, it’s a good value.

Gemini: Best All‑Around Daily Driver

Tier: A

If you just want one $20/month AI that does a bit of everything well, Gemini is arguably the best daily driver right now.

Why it stands out:

• Image generation: Nano Banana Pro is best‑in‑class for images.
• Video generation: Built‑in text‑to‑video that’s competitive with other top tools.
• Video understanding: This is where Gemini really shines. You can feed it YouTube videos, Reels, TikToks, and other clips and it can actually analyze and reason about them. Most other chatbots either struggle here or can’t do it at all.

Model‑to‑model, Gemini 3.1 might not always beat GPT‑5.4 or Claude Opus on pure reasoning, but as a single subscription for the average user, it’s hard to beat.

Claude: Incredible Model, Frustrating Limits

Tier: C for $20/month users, S for $100–$200/month users

Claude’s flagship model, Opus 4.7, is arguably the best LLM available today. For serious builders and power users on the $100 or $200/month plans, it’s S‑tier: amazing reasoning, great coding, and strong long‑form performance.

But as a chatbot on the $20/month plan, it’s a different story:

• You hit usage limits very quickly if you try to use Opus for real projects.
• It only handles text and code—no native images or video in the main artifacts.
• If you’re casual or budget‑constrained, you’ll feel throttled fast.

So for the average person just wanting one chatbot, Claude lands in C tier. For serious users willing to pay for higher tiers, it jumps to S tier.

Grok: Great for Twitter‑Brained Users

Tier: C (B+ if you live on Twitter/X)

Grok’s main selling points are:

• Looser guardrails than the other big models.
• Strong real‑time awareness of what’s happening on X and in the news.

But it falls short elsewhere:

• Weaker at coding than ChatGPT, Claude, or Gemini.
• Image and video generation lag behind competitors.
• Its video handling is really optimized around X/Twitter, not general use.

If you spend four hours a day on Twitter, Grok might feel like B+ tier for you. For everyone else, it’s a C‑tier chatbot you probably wouldn’t recommend as a primary tool.

Coding Agents & Harnesses: Where Real Leverage Lives

This is where things get serious. Coding agents and harnesses are the tools that let you build real software, automations, and AI agents—often without being a traditional developer.

Claude Code: Industry Standard for AI Development

Tier: S (for $100–$200/month users)

Claude Code has quietly become an industry standard for AI‑assisted development and agent building. Paired with Opus, it’s extremely powerful for:

• Building full applications and backends
• Creating complex automations and AI agents
• Handling large codebases and refactors

Important caveat: it’s not realistically usable on the $20/month plan. If you’re serious about building with Claude Code, you’ll want the $100 or $200/month tier. The good news is that even at those prices, you’re still getting a huge discount compared to raw API token costs.

There’s noise online about Claude Code being “nerfed,” but much of that is more about expectations, popularity, and skill gaps than the tool itself. For most builders, it’s still S‑tier.

Codex: OpenAI’s Answer to Claude Code

Tier: S

Codex is OpenAI’s coding harness, essentially their version of Claude Code. It’s also S‑tier and, for many people, interchangeable with Claude Code.

Key strengths:

• Generous usage: You typically get more tokens to work with than on Claude Code.
• Strong models: GPT‑5.4 is more than capable of handling the same kinds of tasks you’d give Opus 4.6 or 4.7. For 99% of users, the difference is marginal.

Choosing between Claude Code and Codex often comes down to personal preference and ecosystem. If you’re wondering which to pick, the honest answer is: pick one and go deep. The skills you build will transfer.

Antigravity: Good, But Held Back by Its Models

Tier: A

Antigravity sits somewhere between a coding harness and an IDE. Not long ago it could have been S‑tier, but today it’s more of an A‑tier option.

The main issue is that when you pair it with Google’s own models like Gemini 3.1, it just doesn’t compete with Claude Code + Opus or Codex + GPT‑5.4. In practice, you’re often better off using Antigravity with those other models—which raises the question: why not just use the native harnesses they’re built for?

Usage complaints are also starting to pop up as the once‑generous limits tighten.

Open Claude & Hermes: Productivity Theater

Open Claude Tier: C (borderline D)
Hermes Tier: C

Open Claude and Hermes are often talked about like coding agents, but they mostly fall into the “performance theater” category:

• Lots of dashboards, heartbeats, and visual activity.
• Feels like a lot is happening—but you could do the same work faster and more reliably inside Claude Code or Codex.

There’s still no compelling use case that must be done in Open Claude and can’t be done better in a proper coding harness. And since Anthropic has restricted using your max subscription inside Open Claude (with only vague claims that this might be changing), it’s even harder to justify.

Hermes adds some clever ideas like self‑updating skills, but that can easily drift into illusion: lots of automatic rewriting without clear evidence it’s actually improving results.

If you’re serious about building, stick to real coding agents.

Cursor: Still Good, But Less Essential

Tier: B (A if you love Composer)

Cursor is an AI‑powered IDE that many people used alongside Claude Code or Codex. A common setup was:

• Use Claude Code to do most of the heavy lifting.
• Use Cursor (with GPT‑5.4) as a second set of eyes for code review.

That workflow is less necessary now, because you can run adversarial code review with Codex inside Claude Code via plugins. If you’re a big fan of Cursor’s Composer 2.0, you might still rate it A‑tier. Otherwise, it’s a solid B‑tier tool: good, but not essential.

Gemini CLI: A Non‑Starter

Tier: D

Gemini CLI exists, but almost no one outside Google seems to actually use it. If a dev tool is basically absent from real‑world workflows, that usually tells you what you need to know.

No‑Code & Low‑Code Builders: Who Still Makes Sense?

No‑code AI builders had a big moment, but many are now struggling to justify their price and existence in a world where Claude Code and Codex can build full apps directly.

Lovable & Bolt: Once Exciting, Now Obsolete

Lovable Tier: C
Bolt Tier: D

Lovable was impressive when it launched: tuned models that generated attractive front‑end designs, one‑click Supabase setups, and easy deployment. Bolt had a similar moment of hype when it could spin up apps from scratch.

Today, the picture is different:

• Claude Code, Codex, and even some chatbots can now generate front‑ends just as good.
• You’re paying a steep premium for convenience that no longer feels unique.
• Any real app still requires you to deal with technical complexity—and if you can handle that, you can just do it in a coding harness.

Lovable isn’t a bad product, but it’s hard to recommend over the competition. Bolt looks even more like a dead end. The same concerns likely apply to similar players like Replit’s AI‑heavy offerings and Base 44 as they try to compete with the big harnesses.

N8N, Make & Zapier: Automation in the Age of Agents

N8N Tier: B (niche but still useful)
Make & Zapier Tier: D

N8N used to be an easy S‑tier recommendation for building AI agents and automations without code. But now:

• Everything N8N can do, Claude Code and Codex can also do.
• You can often spin up N8N‑style workflows faster using Claude Code itself.

So why use N8N at all? It still has a niche:

• Great for non‑technical teams who need to see and tweak automations visually.
• Ideal when clients want a drag‑and‑drop interface they can maintain themselves.

In those cases, N8N’s visual canvas and ease of use justify it as a B‑tier tool.

Make and Zapier, on the other hand, are more expensive, less flexible, and offer little advantage over modern AI harnesses. Outside of legacy setups, they’re hard to recommend.

Research & Knowledge Tools

Notebook LM: The Best Single‑Purpose AI Tool

Tier: S

Notebook LM is arguably Google’s best AI product—and one of the best single‑purpose AI tools overall.

What it’s great for:

• Deep research projects with lots of documents and YouTube videos
• Generating high‑quality slide decks, infographics, and structured notes
• Acting as a “research brain” you can plug into other tools

It integrates nicely with setups involving Claude Code or Antigravity (for example, via the Notebook LM Pi CLI), and Google lets you do a surprising amount for free.

If your work involves heavy research, Notebook LM is S‑tier and pairs extremely well with the kind of workflows covered in guides like advanced Claude Co‑work skill setups.

Perplexity: Great Web Brain, Weak Daily Driver

Tier: B

Perplexity is technically a chatbot, but it’s best thought of as a web‑first AI search engine.

Where it shines:

• Quickly answering questions about things that happened today.
• Surfacing sources and citations without extra prompting.
• Acting like “Grok on steroids” for web search.

Where it falls short:

• It’s not compelling as your only chatbot.
• Paying $20/month on top of other tools is hard to justify for most people unless you live in research mode.

If you constantly need up‑to‑the‑minute information with sources, Perplexity is a strong B‑tier add‑on. Otherwise, it’s a nice‑to‑have, not a must‑have.

Design, Images & Video: Creative AI Tools Ranked

Design by Claude: Brilliant but Crippled by Usage

Tier: B (S‑tier product, D‑tier usage)

Design by Claude is a newer tool focused on rapid visual ideation and UI/UX design. In terms of raw capability, it’s fantastic:

• Huge upgrade over baseline Claude Code for front‑end and visual work.
• Better than most public skills, including older front‑end design skills.
• Great for quickly iterating through multiple visual directions.

The problem is usage:

• Token limits are extremely tight—even on the $200/month Claude plan.
• A single design system can burn ~30% of your allowance.
• A simple landing page can chew through ~5%.

Until the usage model is fixed, it’s hard to recommend as a primary tool, even though the underlying product is excellent.

Video Models: Kling, VEO, and Seed Dance

Kling 3.0 Tier: A
VEO‑3 / VEO‑3.1 Tier: B
Seed Dance 2.0 Tier: S

On the video side:

• Kling 3.0 is a strong A‑tier model with very solid results.
• VEO‑3 / 3.1 is starting to feel dated and is relatively expensive, so it drops to B tier.
• Seed Dance 2.0 is currently leading the pack: it produces wild, cinematic visuals (you’ve probably seen its viral movie‑style clips) and is surprisingly affordable compared to its quality.

If you’re doing serious AI video work, Seed Dance is the one to watch.

Image Models: Nano Banana Pro & GPT Image Gen 2

Nano Banana Pro Tier: S
GPT Image Gen 2 Tier: A (potentially S)

For images:

• Nano Banana Pro (in Gemini) has been best‑in‑class for a while and remains S‑tier.
• GPT Image Gen 2 is very new but already looks excellent. Early use suggests it’s at least A‑tier, with real S‑tier potential once it’s battle‑tested across more edge cases.

A strong image model is a big win for ChatGPT, especially since Gemini’s superior images were one of its main differentiators for a long time.

Copilot: Only When You’re Forced To

Tier: C‑

Copilot (in its various Microsoft flavors) is the kind of tool you use because it’s bundled into your company stack—not because you’d choose it on its own.

It’s fine, and sometimes convenient when deeply integrated into Office or other enterprise tools, but it’s nowhere near the top of the list if you’re picking tools for yourself.

How to Actually Choose: What to Focus On (and What to Ignore)

Looking across all categories, a few patterns emerge:

• S‑tier tools to invest in: Claude Code, Codex, Notebook LM, Seed Dance 2.0, Nano Banana Pro (and likely GPT Image Gen 2), plus Gemini as a daily driver for most people.
• A‑tier tools worth learning: Antigravity, Kling 3.0, GPT Image Gen 2 (for now), and Perplexity if you live in the browser and news cycle.

On the other hand, most tools in the C and D tiers—especially things like Open Claude, Hermes, Lovable, Bolt, Make, and Zapier—are easy ways to burn time and money without gaining skills that transfer.

It’s often better to consolidate around a small stack of high‑leverage tools and go deep. That’s the same mindset behind workflows where people replace 10+ tools with a single AI workspace, like in setups similar to all‑in‑one AI workspaces.

Most importantly, focus on learning the fundamentals of building with AI—software architecture, prompting for complex systems, and how to design robust workflows. If Claude Code or Codex were replaced tomorrow, those skills would carry over to whatever comes next.