Every Claude Model: Complete Guide from Claude 3 to Mythos

The Full Claude Lineage

Anthropic ships fast. In just over two years the Claude family went from a single Claude 3 generation in March 2024 to a deeply tiered Claude 4 lineup, and most recently to the Mythos Preview in April 2026. If you have lost track of which model does what, you are not alone. This guide walks the entire timeline, pricing, context windows, and the decision framework for picking one for daily work.

Timeline of Every Model

Model	Released	Context	Input / Output ($ per Mtok)
Claude 3 Haiku	Mar 2024	200K	$0.25 / $1.25
Claude 3 Sonnet	Mar 2024	200K	$3 / $15
Claude 3 Opus	Mar 2024	200K	$15 / $75
Claude 3.5 Sonnet	Jun 2024	200K	$3 / $15
Claude 3.5 Haiku	Oct 2024	200K	$0.80 / $4
Claude 3.7 Sonnet	Feb 2025	200K	$3 / $15
Claude Sonnet 4	May 2025	200K	$3 / $15
Claude Opus 4	May 2025	200K	$15 / $75
Claude Opus 4.5	Sep 2025	500K	$5 / $25
Haiku 4.5	Oct 2025	500K	$1 / $5
Sonnet 4.6	Dec 2025	1M	$3 / $15
Opus 4.6	Jan 2026	1M	$5 / $25
Opus 4.7	Mar 2026	1M	$5 / $25
Mythos Preview	Apr 2026	1M	$25 / $125

How the Families Differ

The Claude 3 family was built around a clean three-tier split: Haiku for speed, Sonnet for balance, Opus for reasoning. All three sat at 200K tokens of context and used the older Anthropic price band where Opus cost $15 input and $75 output per million tokens. The 3.5 and 3.7 releases improved the middle tier without changing the pricing or the context window.

The Claude 4 family rewrote the playbook. Context jumped from 200K to 500K and then to 1M tokens. Opus dropped from $15/$75 to $5/$25 — three times cheaper at the top tier. Sonnet stayed at $3/$15 but got the full million-token window. Haiku 4.5 became cheap enough at $1/$5 that it is now realistic to route entire workflows through it.

The Mythos Preview is the new top of the stack. At $25 input and $125 output per million tokens it is the most expensive Claude has ever been per token, but the reasoning depth it brings to genuinely hard problems makes it the right tool when the alternative is shipping a bug.

Decision Framework

• Daily driver: Sonnet 4.6 at $3/$15 per million tokens with a 1M token window. The right default for almost everything.
• Heavy reasoning: Opus 4.7 at $5/$25 per million tokens. Architectural decisions, novel debugging, hard refactors.
• Budget tier: Haiku 4.5 at $1/$5 per million tokens. Code review, log analysis, formatting, fast iteration on simple tasks.
• Frontier work: Mythos Preview at $25/$125. Only when Opus 4.7 is not enough — security-critical reasoning, scientific work, complex agentic chains.

What Most Developers Get Wrong

The temptation is to default to the most powerful model available. That is expensive and slower than it needs to be. The other failure mode is defaulting to the cheapest model and burning hours chasing bugs the cheaper model created. The right discipline is matching model to task: route 80 percent of your work to Sonnet 4.6, escalate the 15 percent that needs depth to Opus 4.7, and reach for Mythos only on the rare problem that justifies it.