Migration assistant

Replacing Claude Haiku 4.5?

Ranked replacement candidates, each shown with exactly what changes if you switch. Same-provider matches surface first because switching cost is lower; capability regressions are flagged in red.

AN
Migrating from

Claude Haiku 4.5

Anthropic · Claude Haiku 4 · Active
Input: $1/M Output: $5/M Context: 200K
View detail →
#1 RECOMMENDED

Claude Haiku 3.5

Anthropic · Claude Haiku 3
same provider; 20% cheaper input; same context window; loses 1 capability
View full detail →
What changes if you switch
Field Claude Haiku 4.5 → Claude Haiku 3.5 Impact
Input price $1/M → $0.8/M Save 20%
Output price $5/M → $4/M Save 20%
Context window 200K tokens → 200K tokens Same capacity
Vision input ✓ supported → ✗ not supported LOST — re-evaluate before switching
Function calling ✓ supported → ✓ supported Preserved
Prompt caching ✓ supported → ✓ supported Preserved
Lifecycle Active → Active Same status
Other candidates
#2

Claude Opus 4

Anthropic
same provider; 1400% more expensive input; same context window
Details →
Input price: +1400% more expensiveOutput price: +1400% more expensive
#3

Claude Opus 4.1

Anthropic
same provider; 1400% more expensive input; same context window
Details →
Input price: +1400% more expensiveOutput price: +1400% more expensive
#4

Claude Sonnet 4

Anthropic
same provider; 200% more expensive input; same context window
Details →
Input price: +200% more expensiveOutput price: +200% more expensive
#5

Gemini 2.5 Flash

Google Gemini
70% cheaper input; 1M context (larger)
Details →
Input price: Save 70%Output price: Save 50%Context window: +424% larger
Methodology

How candidates are ranked

Candidates are ranked by how much of the source model's profile each one preserves — weighing switching cost (same provider or model family surface first), capability parity (what you keep versus what you'd lose), context window, and pricing direction. A candidate that is itself deprecated is pushed down the list; one that's already retired is never recommended.

Ranking is purely algorithmic — no editorial weighting, no paid placement. Every value is pulled from each provider's own documentation; click any model name to see the source-linked detail.