GPT-4o Mini vs Claude Haiku vs Gemini Flash: The Budget Model Showdown
Every major AI provider now has a "budget" model. OpenAI has GPT-4o Mini. Anthropic has Claude Haiku. Google has Gemini Flash.
But which one is actually the best value? We analyzed pricing across 80 providers to find out.
The Contenders
Three models dominate the budget tier:
| Model | Provider | Context Window | Key Strength |
|---|---|---|---|
| GPT-4o Mini | OpenAI | 128K | Speed, broad compatibility |
| Claude Haiku 4.5 | Anthropic | 200K | Largest context, coding |
| Gemini 2.0 Flash | 1M+ | Massive context, multimodal |
Direct API Pricing Comparison
Prices per 1 million tokens, direct from each provider:
| Model | Input | Output | Cached Input |
|---|---|---|---|
| Gemini 2.0 Flash | $0.10 | $0.40 | $0.025 |
| GPT-4o Mini | $0.15 | $0.60 | $0.08 |
| Claude 3 Haiku | $0.25 | $1.25 | $0.03 |
| Claude 3.5 Haiku | $0.80 | $4.00 | $0.08 |
| Claude Haiku 4.5 | $1.00 | $5.00 | $0.10 |
Winner on base price: Gemini 2.0 Flash at $0.10 input.
But wait—there's more to this story.
The Gemini Flash Family (It's Confusing)
Google has seven different Flash variants. Here's the breakdown:
| Model | Input | Output | Best For |
|---|---|---|---|
| Gemini 1.5 Flash-8B | $0.0375 | $0.15 | Absolute cheapest |
| Gemini 2.0 Flash-Lite | $0.075 | $0.30 | Budget + speed |
| Gemini 2.0 Flash | $0.10 | $0.40 | General purpose |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | Newer, same price |
| Gemini 2.5 Flash | $0.30 | $2.50 | Latest capabilities |
| Gemini 3 Flash Preview | $0.50 | $3.00 | Cutting edge |
The actual cheapest option: Gemini 1.5 Flash-8B at $0.0375/1M input tokens—4x cheaper than GPT-4o Mini.
Cross-Provider Pricing
The same model costs different amounts depending on where you access it:
GPT-4o Mini Across Providers
| Provider | Input | Output | Notes |
|---|---|---|---|
| OpenAI (direct) | $0.15 | $0.60 | Standard pricing |
| Azure OpenAI | $0.15 | $0.60 | Same as direct |
| OpenRouter | $0.15 | $0.60 | Same as direct |
| Poe | $0.14 | $0.54 | 7% cheaper |
| GitHub Models | $0.00 | $0.00 | Free (rate limited) |
| Helicone | $0.15 | $0.60 | Passthrough |
Claude Haiku 4.5 Across Providers
| Provider | Input | Output | Notes |
|---|---|---|---|
| Anthropic (direct) | $1.00 | $5.00 | Standard pricing |
| Amazon Bedrock | $1.00 | $5.00 | Same as direct |
| Azure OpenAI | $1.00 | $5.00 | Same as direct |
| Google Vertex | $1.00 | $5.00 | Same as direct |
| Poe | $0.85 | $4.30 | 15% cheaper |
| AIHubMix | $1.10 | $5.50 | 10% markup |
| Firmware | $0.00 | $0.00 | Free tier |
Gemini 2.0 Flash Across Providers
| Provider | Input | Output | Notes |
|---|---|---|---|
| Google (direct) | $0.10 | $0.40 | Standard pricing |
| Vertex AI | $0.10 | $0.40 | Same as direct |
| OpenRouter | ~$0.10 | ~$0.40 | Passthrough |
The Hidden Cost: Output Tokens
Here's what most comparisons miss: output tokens cost 4-5x more than input tokens.
If your application is output-heavy (generating content, code, long responses), the output price matters more than input.
Output-heavy workload example (1M input, 2M output):
| Model | Input Cost | Output Cost | Total |
|---|---|---|---|
| Gemini 2.0 Flash | $0.10 | $0.80 | $0.90 |
| GPT-4o Mini | $0.15 | $1.20 | $1.35 |
| Claude 3 Haiku | $0.25 | $2.50 | $2.75 |
| Claude Haiku 4.5 | $1.00 | $10.00 | $11.00 |
Gemini Flash is 12x cheaper than Claude Haiku 4.5 for output-heavy work.
Caching Changes Everything
All three providers offer input caching at steep discounts:
| Model | Standard Input | Cached Input | Savings |
|---|---|---|---|
| Gemini 2.0 Flash | $0.10 | $0.025 | 75% off |
| GPT-4o Mini | $0.15 | $0.08 | 47% off |
| Claude Haiku 4.5 | $1.00 | $0.10 | 90% off |
Claude's cache discount is the steepest. If you can structure prompts to maximize cache hits, Claude Haiku 4.5's effective cost drops dramatically.
Cached input comparison:
| Model | Cached Input |
|---|---|
| Gemini 2.0 Flash | $0.025 |
| Claude 3 Haiku | $0.03 |
| GPT-4o Mini | $0.08 |
| Claude Haiku 4.5 | $0.10 |
With caching, Gemini and Claude 3 Haiku are nearly identical.
Which Should You Choose?
Choose Gemini 2.0 Flash if:
- Cost is your primary concern
- You need massive context windows (1M+ tokens)
- You're doing multimodal work (images, video)
- Output generation is heavy
Choose GPT-4o Mini if:
- You need broad ecosystem compatibility
- You're already on OpenAI's platform
- You need the fastest response times
- You want free access via GitHub Models
Choose Claude Haiku if:
- You need the best coding performance in the budget tier
- 200K context is sufficient
- You can leverage aggressive caching (90% discount)
- You need Claude's safety/style for your use case
The Surprise Winner
For pure cost optimization: Gemini 1.5 Flash-8B at $0.0375 input.
For best balance of capability and cost: Gemini 2.0 Flash at $0.10 input.
For coding tasks: Claude 3.5 Haiku at $0.80 input (not the newer 4.5).
For ecosystem and speed: GPT-4o Mini at $0.15 input.
Real-World Cost Calculator
Processing 10 million tokens per month (5M in, 5M out):
| Model | Monthly Cost |
|---|---|
| Gemini 1.5 Flash-8B | $0.94 |
| Gemini 2.0 Flash | $2.50 |
| GPT-4o Mini | $3.75 |
| Claude 3 Haiku | $7.50 |
| Claude 3.5 Haiku | $24.00 |
| Claude Haiku 4.5 | $30.00 |
The gap between cheapest and most expensive: 32x.
All pricing data from Subquery's real-time database covering 80 providers. Compare models →