What AI Model Should I Use? The 2026 Guide to ChatGPT, Claude, Gemini, and Grok

Claude Fable 5 is back. On July 1, 2026, Anthropic restored global access to its most capable publicly available model, 18 days after a US government export-control directive forced it offline — the first time a frontier AI model has been switched off, and back on, by regulatory order. The US Commerce Department lifted the controls on June 30, and Fable 5 returned across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork, wearing a stricter new safety classifier.

The comeback landed in the same week Anthropic reshuffled its entire lineup. On June 30, the company launched Claude Sonnet 5 — a cheaper mid-tier model that gets close to flagship performance — and made it the new default for Free and Pro users. In one week, the Claude hierarchy changed at both the top and the middle.

Which makes this a good moment to answer the question people actually type into search bars: what AI model should I use? The honest answer has changed shape in 2026, and it is now less about picking a brand and more about understanding how every brand’s lineup works. This guide covers all four major assistants — ChatGPT, Claude, Gemini, and Grok — their sub-models, their pricing, and which one fits which person.

TL;DR: The short answers

• Most people should start free. ChatGPT, Claude, Gemini, and Grok all have capable free tiers, and Gemini’s free tier is currently the most generous.
• For coders, Claude remains the developer favorite: Sonnet 5 for everyday work, Opus 4.8 or Fable 5 for the hardest problems.
• For people who live in Gmail, Docs, and Drive, Gemini is the obvious pick — no other assistant is wired that deeply into an ecosystem you already use.
• For heavy writers and knowledge workers, Claude Sonnet 5 and ChatGPT (GPT-5.5) are the strongest all-round choices at $20/month.
• For real-time news, social sentiment, and market chatter, Grok is unmatched because of its live X and web search.
• For students on a budget, Gemini free or Google AI Plus (~$7.99/month) delivers the most capability per dollar.
• For developers watching API bills, Grok 4.1 Fast ($0.20/$0.50 per million tokens) and Gemini Flash-Lite ($0.25/$1.50) are the cheapest credible options; Claude Sonnet 5 at its introductory $2/$10 is the best value at the quality frontier.
• The general rule for 2026: match the assistant to your ecosystem and task, then match the effort level to the difficulty of each job. That second step is new — and most people skip it.

Why “which AI?” got harder and easier at the same time

Two structural shifts define the 2026 model landscape, and once you see them, every product announcement makes more sense.

Shift one: every lab now ships a tiered family, not a flagship

There is no longer one “ChatGPT model” or one “Claude.” Every lab sells a range: a big expensive model for the hardest problems, a mid-sized workhorse, and a cheap fast model for volume. Anthropic ships Fable 5, Opus 4.8, Sonnet 5, and Haiku 4.5. OpenAI’s new generation is Sol, Terra, and Luna. Google runs Pro, Flash, and Flash-Lite. xAI has Grok 4 Heavy, Grok 4.3, and Grok 4.1 Fast.

The strategic logic flipped in 2026, too. The old game was “biggest model wins.” The new game is best cost-performance — and the clearest evidence is that two of the four labs now route their default consumer traffic to a cheaper, near-flagship model rather than the top of the range. Google made Gemini 3.5 Flash its consumer default in May; Anthropic made Sonnet 5 the default for Free and Pro plans on June 30. The flagships still exist, but they have become the exception you reach for, not the baseline you get.

Shift two: the effort dial

The second shift is subtler and matters more for your results. In 2025, if you wanted more intelligence, you switched models. In 2026, you increasingly turn a dial on the same model.

What is “effort”? In plain English: how long and hard the model is allowed to think before answering. At low effort, the model responds quickly and cheaply. At high effort, it spends more computation reasoning through the problem — checking its work, exploring alternatives — which produces better answers on hard tasks and costs more (in tokens, time, or usage allowance). It is roughly the difference between asking a colleague for a quick take versus asking them to go away and really work the problem.

Every lab now exposes some version of this dial. Claude has effort levels (low, medium, high, and extra-high) available on all claude.ai plans next to the model selector. Gemini has thinking levels and a “Deep Think” mode. Grok 4.3 ships with reasoning on by default and a configurable reasoning_effort in the API. OpenAI’s GPT-5.6 Sol adds a “max” reasoning setting and an “ultra” mode that farms work out to subagents.

The practical consequence: “which model should I use?” has quietly become “which model, and at what effort?” A mid-tier model at high effort can match a flagship at medium effort — Anthropic’s own charts show Sonnet 5 at extra-high effort performing roughly in line with Opus 4.8 on some agentic benchmarks. Sometimes the mid-tier model run that hard actually costs more, which is why the dial is now a genuine cost lever, not a gimmick.

Here’s how to choose an AI model, and the effort level using Claude

The four families at a glance

As of July 2026. Model lineups change monthly; treat this as a snapshot.

The Claude spotlight: what just changed

Is Claude Fable 5 back?

Yes. Claude Fable 5 returned to users globally on July 1, 2026, across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork, after the US Commerce Department lifted the export controls it imposed on June 12. Availability on AWS, Google Cloud, and Microsoft Foundry is following “as quickly as possible,” per Anthropic.

The suspension is worth understanding because it is the strangest product story of the year. Fable 5 launched on June 9 as the first Mythos-class model made generally available — Anthropic’s most capable public model, sitting above Opus 4.8, with roughly a 1-million-token context window and strengths in software engineering, knowledge work, vision, and long-running autonomous tasks. Three days later, after Amazon researchers reported a prompting technique that bypassed some of its cyber safeguards, the Commerce Department ordered Anthropic to cut off access for any foreign national — a condition the company could not verify in real time, so it took the model dark for everyone, worldwide. BNC covered the shutdown as it happened in Washington Pulls the Plug on Anthropic’s Most Powerful AI.

Two details matter for anyone choosing a model today. First, Fable came back slightly different. Anthropic trained a new safety classifier that blocks the reported jailbreak technique in over 99% of cases — but it also flags more benign requests during routine coding and debugging. Blocked requests are automatically rerouted to Opus 4.8, with a notification. If you are doing security-adjacent development work, expect occasional friction. Second, the access terms are time-boxed. For Pro, Max, Team, and select Enterprise plans, Fable 5 is included for up to 50% of weekly usage limits through July 7, 2026, after which it moves to usage credits. In other words: included access is a one-week window; after that, Fable becomes a metered premium.

The Claude hierarchy, in order

Anthropic now sells four generally available models plus one you can’t have:

• Claude Fable 5 — the most capable publicly available model. Best for the hardest agentic, coding, and research tasks. API pricing sits at a premium above Opus; on subscriptions it consumes usage credits after July 7. Reach for it when the task genuinely justifies the cost.
• Claude Opus 4.8 — the previous flagship (launched May 28, 2026), now second in the lineup. API: $5 input / $25 output per million tokens. Still the pick when small accuracy differences are expensive — and the model Fable falls back to when its classifier triggers.
• Claude Sonnet 5 — the new default for Free and Pro plans (launched June 30). Introductory API pricing of $2/$10 per million tokens through August 31, 2026, then $3/$15. It scores 63.2% on SWE-bench Pro versus Opus 4.8’s 69.2%, and actually edges Opus on a knowledge-work benchmark. Anthropic’s framing is honest: near-flagship performance at roughly 40% of the flagship’s per-token cost.
• Claude Haiku 4.5 — the cheap, fast tier at $1/$5. It is also the oldest model in the current lineup, and it shows on harder tasks. Fine for classification, extraction, and high-volume simple work.
• Claude Mythos 5 — Fable’s less-safeguarded sibling, built for offensive-grade cybersecurity work. Not for general users: access was restored on June 26 to a set of vetted US organizations in Anthropic’s Project Glasswing program, with broader access still being negotiated. BNC unpacked the unusual two-product structure in Anthropic Splits the AI Frontier in Two.

Why a cheaper default matters

Making Sonnet 5 the default is not a downgrade dressed up as a launch — for most users it is a straight upgrade over the previous default, and it signals where the economics are heading. When the mid-tier model closes most of the gap to the flagship, the rational default for a lab serving hundreds of millions of requests is the cheaper one. Users get near-flagship answers; the lab saves an enormous amount of compute; the flagship becomes a deliberate choice rather than an ambient cost.

One caveat worth knowing if you pay per token: Sonnet 5 uses an updated tokenizer, and the same text can map to roughly 1.0–1.35× more tokens than under Sonnet 4.6. The rate card looks flat at $3/$15 after August, but your actual per-task cost can drift higher than the sticker suggests. BNC’s launch coverage has the full breakdown: Anthropic Launches Claude Sonnet 5.

The four assistants in depth

ChatGPT (OpenAI): the broadest all-rounder

What it is: ChatGPT is OpenAI’s consumer assistant, currently running the GPT-5 family with an auto-router that picks between Instant and Thinking modes (plus a Pro mode on top tiers). The current consumer flagship is GPT-5.5; the free tier runs the older GPT-5.3 Instant with a small context window.

The GPT-5.6 situation: On June 26, 2026, OpenAI previewed GPT-5.6 — a three-model family named Sol (flagship), Terra (balanced, roughly GPT-5.5 quality at about half the cost), and Luna (fast and cheap). The catch: at the US government’s request, the launch is a limited preview for roughly 20 vetted partner organizations, available only via the API and Codex. GPT-5.6 is not in the consumer ChatGPT app yet. OpenAI says general availability is coming “in the coming weeks,” but as of early July there is no date, no waitlist, and no way to buy your way in. If you subscribe to ChatGPT today, you are buying GPT-5.5 — which is still an excellent model — not 5.6. BNC covered the gated launch in GPT-5.6 Launches Under Government Watch.

Signature strengths: breadth. ChatGPT has the most complete feature surface of the four: best-in-class voice mode, ChatGPT Images 2.0, Sora video generation, the Codex coding agent, Agent Mode for autonomous tasks, and 60+ connectors into third-party apps. If you want one assistant that does a bit of everything well, this is it.

Weak spots: the model-picker experience is opaque (the auto-router decides for you), the free tier is the stingiest of the four on model quality, and the top-end Pro subscriptions are expensive. Developers who need frontier capability today are stuck on 5.5 pricing while 5.6 sits behind the government gate.

Claude (Anthropic): the professional’s coding and writing tool

What it is: Claude is Anthropic’s assistant, covered in depth above. The short version of its character: Claude models are consistently the developer community’s favorite for real-world software engineering and long agentic tasks, and Sonnet-class models have been the “workhorse” pick for professional writing and analysis for two years running.

Signature strengths: coding and agentic reliability — finishing multi-step tasks rather than stalling partway — plus long-document work, careful writing, and (with Fable 5) the strongest publicly available model on several benchmarks. Claude Code has become the default command-line coding agent for a large slice of professional developers.

Weak spots: no image or video generation to speak of, a smaller consumer feature set than ChatGPT or Gemini, and — a genuinely new consideration — regulatory exposure. Fable’s 18-day suspension demonstrated that access to the top of Anthropic’s lineup can be interrupted by policy. The new classifier’s false positives on benign coding work are a real, if minor, tax. And Haiku 4.5, the budget tier, is overdue for a refresh.

Gemini (Google): the ecosystem play with the biggest context

What it is: Gemini is Google’s assistant, woven into Search (AI Mode), Gmail, Docs, Drive, Calendar, Android, and Chrome. The current family: Gemini 3.1 Pro (the flagship reasoning model, with the largest context windows of the four families — up to 2 million tokens on Pro-class models), Gemini 3.5 Flash (the newest model and the consumer default since May 19, 2026), 3.1 Flash-Lite (budget), and 2.5 Deep Think (extended reasoning, reserved for top tiers).

Here is one of the more telling facts of 2026: Google shipped the Flash tier of its 3.5 generation before the matching Pro — and 3.5 Flash beats the older 3.1 Pro on coding and agentic benchmarks at about a quarter of the latency. The cheap model outrunning last generation’s flagship is the tiered-family logic taken to its conclusion. The matching Gemini 3.5 Pro (2M context, Deep Think) was targeted for June but has slipped to July 2026, so for now 3.1 Pro remains the flagship you can actually use.

Signature strengths: if your life runs on Google, nothing else comes close — Gemini reads your Gmail, drafts in your Docs, and lives in your Search results. Add the largest context windows on the market, Veo video generation, NotebookLM, and a genuinely strong Deep Research mode, and the free tier is arguably the best in the business.

Weak spots: the model naming and plan structure are confusing even by AI-industry standards, prices have been changing quickly (Ultra was cut from $249.99 to a $99.99 entry point at I/O 2026), and the very best reasoning (Deep Think) is locked behind the most expensive tiers. Google also has a habit of announcing ahead of shipping — see the 3.5 Pro slip.

Grok (xAI): real-time data at ruthless prices

What it is: Grok is xAI’s assistant, distinguished by live access to X (Twitter) and the web. The flagship is Grok 4.3 (since April 30, 2026): always-on reasoning, a 1M-token context window, native video input, and startlingly low API pricing at $1.25/$2.50 per million tokens. Above it sits Grok 4 Heavy, a multi-agent system that runs up to 16 parallel agents on a single problem, exclusive to the $300/month top tier. Below it, Grok 4.1 Fast offers a 2M context window at $0.20/$0.50 — one of the cheapest credible frontier-adjacent APIs anywhere. A dedicated coding model, grok-build, rounds out the family.

Signature strengths: real-time. DeepSearch pulls live X and web data into cited research reports, which makes Grok the strongest of the four for breaking news, market sentiment, and anything where “what are people saying right now” is the actual question. The API pricing is genuinely disruptive.

Weak spots: two, and both are about transparency. First, xAI bills separately for tools — web search, X search, and code execution each cost about $5 per 1,000 calls on top of tokens, so agentic workloads cost meaningfully more than the token price implies. Second, staged rollouts make it hard to know which model you are actually getting: as of mid-2026, only the $300/month SuperGrok Heavy tier has confirmed full Grok 4.3 access, while cheaper tiers receive it in stages with no indicator of which variant answered your query. Grok’s base training data is also older than rivals’, which it papers over with live search — fine for current events, weaker for offline reasoning.

What does each AI cost in 2026?

Consumer subscription pricing

Prices as of July 2026, US dollars, monthly. Verify before buying — several of these changed within the last six weeks.

Value verdicts: The $20 tier is where most people should land, and all four are defensible there. Gemini AI Pro is the best pure value (1M context, 2TB storage, Veo, Deep Research); Claude Pro is the best for work output; ChatGPT Plus is the best generalist. Below $20, Gemini’s free tier and AI Plus beat everyone. Above $20, be honest with yourself: the $100+ tiers (ChatGPT Pro, Claude Max, Gemini Ultra, SuperGrok Heavy) only pay for themselves if you are hitting usage limits weekly or need a gated capability (Deep Think, Grok 4 Heavy, high Fable usage). What to avoid overpaying for: a top tier bought “just in case,” and X Premium+ at $40 if you only want Grok — standalone SuperGrok at $30 gives you more AI for less.

Developer API pricing

USD per million tokens, input/output, standard rates, July 2026.

Which is cheapest for developers? On raw token price, Grok 4.1 Fast and Gemini Flash-Lite. But cheapest-per-token is not cheapest-per-task: Grok’s tool fees add real cost to agentic workloads, Gemini’s Pro rates double past 200K input tokens, and Sonnet 5’s denser tokenizer means identical text bills more tokens than it did on Sonnet 4.6. The best value at the quality frontier right now is Claude Sonnet 5 at its introductory $2/$10 — a window that closes August 31 — with Gemini 3.5 Flash and GPT-5.6 Terra (once it actually ships to everyone) close behind. Benchmark your own workload; the effort dial means two teams using the same model can see wildly different bills.

Which AI model should you use?

Match the assistant to your ecosystem and dominant task first; the table below covers the common cases.

Then apply the second half of the rule: match effort to difficulty. Run everyday tasks at default or medium effort. Save high and extra-high effort — and flagship models — for the tasks where a wrong answer is expensive. Most people overthink the brand choice and never touch the dial.

What people get wrong about choosing an AI model

“The most expensive model is the best choice.” Usually false. Sonnet 5 at medium effort is the better buy than Opus 4.8 for the majority of tasks; Gemini 3.5 Flash beats the pricier 3.1 Pro on coding. Pay up only when the task demands it.

“The free tiers are toys.” No longer true. Gemini’s free tier runs the same 3.5 Flash model that powers Google’s paid defaults, and Claude’s free tier now serves Sonnet 5. The free/paid gap in 2026 is mostly about usage limits and extras, not raw intelligence.

“A cheaper price per token means a cheaper bill.” Not necessarily. Grok bills roughly $5 per 1,000 tool calls on top of tokens; Gemini Pro rates double past 200K input tokens; Sonnet 5’s updated tokenizer turns the same text into up to ~1.35× more tokens. Cost per completed task is the only number that matters.

“You’re always getting the model on the label.” Also not guaranteed. ChatGPT’s auto-router picks modes for you; Grok’s staged rollouts mean two subscribers on the same plan can hit different model versions; Fable 5 silently falls back to Opus 4.8 when its safety classifier triggers.

A note on the governance backdrop

One genuinely new factor belongs in any 2026 buying decision: frontier models can now be switched off by policy. Fable 5’s 18-day, government-ordered suspension and GPT-5.6’s launch behind a government-managed access list — the first frontier model released that way — happened within a fortnight of each other. The interruption also handed measurable momentum to rivals, including Chinese open-weight models such as GLM-5.2, which briefly inherited benchmark top spots while Fable was dark. None of this makes any single assistant a bad choice. It does mean that if AI access is operationally critical to you, treat sudden model unavailability as a real risk and keep a second assistant warm. This guide takes no view on whether the interventions were wise — only that they happened, and can happen again.

How to choose: the simple rule

If you remember one thing from this guide, make it this two-step rule:

1. Match the assistant to your ecosystem and your dominant task. Google-centric life → Gemini. Serious coding or writing → Claude. A bit of everything, plus voice and media → ChatGPT. Real-time information → Grok.
2. Match the effort to the difficulty. Default effort for daily work; high effort and big models for the few tasks that genuinely deserve them.

Start free, upgrade to the $20 tier when you hit limits, and re-read the pricing tables before you buy anything above that. And date-check everything: in a year when a flagship model was turned off and on again by government order inside three weeks, the only safe assumption is that this landscape will look different by spring.

FAQ

Is Claude Fable 5 back?

Yes. Fable 5 returned globally on July 1, 2026, on Claude.ai, the Claude Platform, Claude Code, and Claude Cowork after the US lifted export controls on June 30. For Pro, Max, Team, and select Enterprise plans it is included for up to 50% of weekly usage limits through July 7, 2026, then available via usage credits.

Is Fable 5 better than Opus 4.8?

Yes, on capability — Fable 5 is Anthropic’s most capable publicly available model and outscores Opus 4.8 on multiple benchmarks. But Fable carries stricter cyber/bio safeguards (with more false positives on benign coding since its redeployment) and costs more to use, so Opus 4.8 remains the practical pick for many demanding tasks.

What’s the difference between Fable, Opus, Sonnet, and Haiku?

They are tiers of the Claude family, from most to least capable: Fable 5 (top public model, premium price), Opus 4.8 (previous flagship, $5/$25 per million tokens), Sonnet 5 (near-flagship workhorse and the default model, $2–3/$10–15), and Haiku 4.5 (cheap and fast, $1/$5). Mythos 5 is a restricted cyber-focused sibling of Fable, available only to vetted organizations.

What AI model should I use in 2026?

For most people: the free or $20 tier of the assistant that matches their ecosystem — Gemini for Google users, Claude for coders and writers, ChatGPT for generalists, Grok for real-time information — then adjust the effort level to the task rather than defaulting to the biggest model.

Which AI is best for free?

Gemini. Its free tier runs Gemini 3.5 Flash — the same model behind Google’s paid defaults — with image generation, voice, and a monthly Deep Research allowance. Claude’s free tier (Sonnet 5) is the strongest free option for writing and coding quality.

Which AI is cheapest for developers?

On raw token prices, Grok 4.1 Fast ($0.20/$0.50 per million tokens) and Gemini 3.1 Flash-Lite ($0.25/$1.50). At the quality frontier, Claude Sonnet 5’s introductory $2/$10 (through August 31, 2026) is the standout value. Always model total cost per task: tool fees, long-context surcharges, and tokenizer differences can outweigh the rate card.

Which AI is best for students?

Gemini — the free tier plus NotebookLM covers most study workflows, and Google AI Plus (~$7.99/month) is the cheapest meaningful paid upgrade of the four families. Claude free is the best companion for essay drafting and coding coursework.

Which AI is best for coding?

Claude. Sonnet 5 is the best value coding model available right now, Opus 4.8 and Fable 5 lead on the hardest agentic engineering benchmarks, and Claude Code is the most widely adopted coding agent. Gemini 3.5 Flash and (once broadly released) GPT-5.6 Sol are the strongest challengers.

---

This guide is updated as the model landscape changes. Figures verified as of July 2, 2026.