We build with it. Then we review it.
Best & Tested. Hands-on tool reviews · Tested 2026
HomeAI Models › Claude AI Models Explained (2026): Opus 5, Fable 5, Sonnet 5 & Haiku Compared
AI Models The Ranking

Claude AI Models Explained (2026): Opus 5, Fable 5, Sonnet 5 & Haiku Compared

Anthropic shipped four models in under two months, and the naming is a maze. Here is the plain-English map: what each Claude tier is actually for, what it costs, and which one is doing the work inside the AI app builders we review.

Claude AI Models Explained (2026): Opus 5, Fable 5, Sonnet 5 & Haiku Compared
On this page (6 sections)
Quick answer

Which Claude model should I use in 2026? For most people and most work, Claude Opus 5 — released 24 July 2026, it lands within a whisker of Anthropic's flagship on coding and agent tasks while costing half as much ($5 vs $10 per million input tokens). Step up to Claude Fable 5 only for the hardest long-horizon autonomous work, drop to Sonnet 5 when you want near-Opus quality at high volume, and use Haiku 4.5 for fast, cheap, simple jobs.

Key takeaways
  1. Claude Opus 5 (Jul 2026) is the new sweet spot: near-flagship intelligence at $5/$25 per million tokens.
  2. Claude Fable 5 is the flagship — Anthropic's most capable model, priced at $10/$50 for the ceiling on agentic work.
  3. Claude Sonnet 5 ($3/$15) is the volume workhorse; Haiku 4.5 ($1/$5) is the speed-and-cost pick.
  4. All the top models carry a 1M-token context window and 128K max output.
  5. This matters for builders: Cursor, Bolt, Replit and Lovable all lean on Anthropic's Claude models — the tier you pick changes how well they reason.

at a glance

Models coveredOpus 5, Fable 5, Sonnet 5, Haiku 4.5 (+ Mythos 5)
NewestClaude Opus 5 — released 24 Jul 2026
Best valueClaude Opus 5 — $5 / $25 per million tokens
Most capableClaude Fable 5 — $10 / $50 per million tokens
Context window1M tokens on every top model

Anthropic spent the first half of 2026 releasing models faster than most people could keep track of them. Fable 5 arrived in June. Sonnet 5 followed weeks later. Then Opus jumped from 4.8 straight to Opus 5 on 24 July. The result is a lineup with genuinely different tiers and a naming scheme that hides which one you should actually reach for.

This guide fixes that. Every number below is pulled from Anthropic’s own documentation, and the framing comes from the thing we care about on this site: these are the models running inside the AI app builders we test, so picking the right one changes what those tools can do.

4main models in the lineup
$1–$10input price range per M tokens
1Mtoken context on the top tiers

The 2026 Claude lineup at a glance

Four models cover the spread from “cheap and instant” to “most capable model Anthropic ships.” Prices are per million tokens, input / output.

ModelBest forContextPrice (in/out)
Opus 5 Everyday best value — coding & agents1M$5 / $25
Fable 5 The absolute ceiling; long-horizon agents1M$10 / $50
Sonnet 5 Near-Opus quality at high volume1M$3 / $15
Haiku 4.5 Fast, cheap, simple tasks200K$1 / $5

A few things worth knowing that the table can’t show. Fable 5 always runs with thinking on and is the only widely released model Anthropic calls “most capable.” Opus 5 is the newcomer that reset the value equation — same price as the old Opus 4.8, but Anthropic says it roughly doubles that model’s coding score. Sonnet 5 is running introductory pricing of $2/$10 through 31 August 2026 before it settles at $3/$15. And Mythos 5, which you’ll see mentioned, is a Fable 5 twin reserved for approved security customers through Project Glasswing — not something you can sign up for.

Which Claude model should you use?

Skip the benchmark tables and answer one question: what are you optimizing for?

MOST PEOPLE

Opus 5

The default. It gets within roughly 0.5% of the flagship on coding evals at half the cost, with a 1M context window. Unless you have a reason to pay more or less, start here.

Opus 5 explained →
THE CEILING

Fable 5

For the hardest autonomous, long-running agent work where the last few percent of capability is worth double the price. Otherwise it is overkill.

Fable 5 explained →
HIGH VOLUME

Sonnet 5

When you are making a lot of calls and want most of Opus quality at a lower rate. The workhorse for production apps and chat features.

See the lineup →

For anything latency-sensitive or high-throughput and simple — classification, quick summaries, routing — Haiku 4.5 at $1/$5 is the cheapest way to get near-frontier quality without waiting.

Opus 5 vs Fable 5: the comparison that matters

This is the decision most people are actually trying to make, so it’s worth being precise. Fable 5 is the flagship and holds the top of Anthropic’s own charts. Opus 5 is the one that launched in July at half the input price. On Anthropic’s coding benchmarks the gap is small: Opus 5 comes within about half a percent of Fable 5 on CursorBench, and on a couple of agent evaluations it actually wins. It also doubles the previous Opus generation on Anthropic’s Frontier-Bench coding test.

The honest read: pay for Fable 5 when you’re running the most demanding autonomous agents and the ceiling genuinely pays for itself. Use Opus 5 for everything else. We break both down in full in the Claude Opus 5 deep-dive and the Claude Fable 5 deep-dive.

How Claude powers the tools you actually build with

Here’s why any of this matters if you never open the API. The AI app builders we review don’t have their own intelligence — they call a frontier model, and for a lot of them that model is Claude. Agentic editors like Cursor and Bolt let you choose it outright, and prompt-to-app builders such as Lovable and Replit lean on Anthropic’s models to turn your prompt into working software.

So the model generation is a lever on your results. A builder pointed at Opus 5 reasons about your codebase very differently than one on an older model. The fastest way to put a current Claude model to work — without touching a single line of setup — is inside a builder that already runs on it.

Pricing, side by side

ModelInput / MOutput / MMax output
Haiku 4.5 $1$564K
Sonnet 5 $3 (intro $2)$15 (intro $10)128K
Opus 5 $5$25128K
Fable 5 $10$50128K

Prefer flat pricing to per-token billing? The chat apps bundle the top models: Claude Pro for individuals and Claude Max for heavy users, with Opus 5 as the new default on Max.

The bottom line

The lineup is less confusing than the names suggest. Opus 5 is the model most people should use — it reset the price-to-capability ratio in July 2026 and covers the vast majority of real work. Reach past it to Fable 5 only for the hardest agent tasks, drop to Sonnet 5 for volume, and keep Haiku 4.5 around for the fast, cheap jobs. And if you’re building rather than calling the API directly, the practical move is to run one of these models inside an app builder we’ve tested.

Frequently asked questions

What are all the Claude models in 2026?
Anthropic's current lineup is Claude Fable 5 (the flagship), Claude Opus 5 (released July 2026), Claude Sonnet 5, and Claude Haiku 4.5. There is also Claude Mythos 5, which shares Fable 5's specs but is limited to approved customers through Project Glasswing. Older Opus 4.x and Sonnet 4.x models are still available but no longer the default.
Is Claude Opus 5 better than Fable 5?
Not quite — Fable 5 remains Anthropic's most capable model for the most demanding autonomous work. But Opus 5 is close. On coding benchmarks like CursorBench it sits within about 0.5% of Fable 5 while costing half as much, and on some agent tasks it comes out ahead. For most people Opus 5 is the smarter buy; Fable 5 is for when you genuinely need the ceiling.
How much do Claude models cost?
API pricing per million tokens (input / output): Haiku 4.5 is $1/$5, Sonnet 5 is $3/$15, Opus 5 is $5/$25, and Fable 5 is $10/$50. If you use Claude through the chat apps instead, Claude Pro and Claude Max bundle the top models into a flat monthly fee.
Which Claude model do AI app builders use?
It varies by tool and often by plan. Agentic editors like Cursor let you pick the model directly, and Anthropic's Claude models are the popular choice for coding. Prompt-to-app builders such as Lovable and Replit lean on Claude under the hood. That's why the model generation matters even if you never touch the API.
What is the context window on Claude models?
Fable 5, Opus 5 and Sonnet 5 all carry a 1M-token context window (roughly 555,000 words) with up to 128K tokens of output. Haiku 4.5 is smaller at 200K tokens in and 64K out. A million tokens is enough to hold a large codebase or a book-length document in a single prompt.
Sources & data
Review changelog

Jul 2026 — Published after the Claude Opus 5 launch; full lineup mapped

M

Marios K.

I build SaaS products and websites for a living. Every tool here is put through the same real-world brief and scored on a fixed rubric — never reviewed from its landing page.

Keep reading