#seedling #note #ai #tool >[!info] >This entire note was written by Claude. I was experimenting with Claude sub-agents. 3 sub-agents were spun, one for each LLM provider. A comparison of pricing models across the three major LLM providers to help determine the most cost-effective option for data processing tasks. ## Quick Comparison - Popular Models (December 2025) | Provider | Model | Input (per 1M tokens) | Output (per 1M tokens) | Context Window | | ------------- | --------------------- | --------------------- | ---------------------- | -------------- | | **OpenAI** | GPT-4o | $5.00 | $15.00 | 128K | | | GPT-4o-mini | $0.15 | $0.60 | 128K | | | o1-mini | $3.00 | $12.00 | 128K | | **Anthropic** | Claude Opus 4.5 | $5.00 | $25.00 | 200K | | | Claude Sonnet 4.5 | $3.00 | $15.00 | 200K | | | Claude Haiku 4.5 | $1.00 | $5.00 | 200K | | | Claude Haiku 3.5 | $0.80 | $4.00 | 200K | | | Claude Haiku 3 | $0.25 | $1.25 | 200K | | **Google** | Gemini 2.5 Pro | $1.25 | $10.00 | 1M | | | Gemini 2.5 Flash | $0.15 | $0.60 / $3.50* | 1M | | | Gemini 2.5 Flash-Lite | $0.10 | $0.40 | 1M | | | Gemini 2.0 Flash-Lite | $0.075 | $0.30 | 1M | *Gemini 2.5 Flash output pricing: $0.60/1M for standard output, $3.50/1M when using reasoning/thinking mode. ## Budget Models Ranked (Cheapest First) For cost-sensitive workloads, these are the most economical options: | Rank | Model | Input | Output | Combined Cost* | |------|-------|-------|--------|----------------| | 1 | Gemini 2.0 Flash-Lite | $0.075 | $0.30 | $0.375 | | 2 | Gemini 2.5 Flash-Lite | $0.10 | $0.40 | $0.50 | | 3 | GPT-4o-mini | $0.15 | $0.60 | $0.75 | | 3 | Gemini 2.5 Flash | $0.15 | $0.60 | $0.75 | | 5 | Claude Haiku 3 | $0.25 | $1.25 | $1.50 | *Combined cost = Input + Output per 1M tokens (assuming equal input/output). Note: most real workloads have asymmetric I/O ratios. ## Cost Estimate: Processing 1 Million Data Records ### Assumptions - Average record size: ~200 tokens (based on a typical structured record with 10-20 fields containing short text, numbers, and categorical values) - Total input: **200M tokens** (1M records × 200 tokens) - Output per record: ~50 tokens (classification label, extracted fields, or brief summary) - Total output: **50M tokens** (1M records × 50 tokens) Token counts vary significantly by data type. A sparse CSV row might be 50 tokens; a JSON object with nested fields could exceed 500. Test with a sample of your actual data. ### Estimated Costs | Model | Input Cost | Output Cost | **Total Cost** | |-------|-----------|-------------|----------------| | **Gemini 2.0 Flash-Lite** | $15.00 | $15.00 | **$30.00** | | **Gemini 2.5 Flash-Lite** | $20.00 | $20.00 | **$40.00** | | **GPT-4o-mini** | $30.00 | $30.00 | **$60.00** | | **Gemini 2.5 Flash** | $30.00 | $30.00 | **$60.00** | | **Claude Haiku 3** | $50.00 | $62.50 | **$112.50** | | **Claude Haiku 3.5** | $160.00 | $200.00 | **$360.00** | | **Claude Haiku 4.5** | $200.00 | $250.00 | **$450.00** | | **Gemini 2.5 Pro** | $250.00 | $500.00 | **$750.00** | | **Claude Sonnet 4.5** | $600.00 | $750.00 | **$1,350.00** | | **GPT-4o** | $1,000.00 | $750.00 | **$1,750.00** | | **Claude Opus 4.5** | $1,000.00 | $1,250.00 | **$2,250.00** | ## Recommendations ### For Maximum Cost Savings **Gemini 2.0 Flash-Lite** or **Gemini 2.5 Flash-Lite** — Processing 1M records for $30-40 ### For Best Value (Cost vs Capability) **GPT-4o-mini** or **Gemini 2.5 Flash** — Both at $60 for 1M records. GPT-4o-mini has mature tooling and documentation; Gemini 2.5 Flash offers a 1M context window. ### For Complex Reasoning on a Budget **Claude Haiku 3** at $112.50, or **Gemini 2.5 Flash with reasoning mode** (output costs increase to $3.50/1M) ### Additional Cost-Saving Strategies 1. **Batch Processing**: OpenAI, Anthropic, and Google all offer ~50% discount for non-urgent batch jobs processed within 24 hours 2. **Prompt Caching**: All three providers offer caching — OpenAI (50% off, automatic), Anthropic (up to 90% off reads), Google (up to 90% off) 3. **Context Optimisation**: Reduce token count by trimming unnecessary fields from records ## Important Notes - Prices are as of December 2025 and subject to change - Actual token counts vary based on data content and encoding - Quality and accuracy vary by model — test on a representative sample before committing to a provider - Consider rate limits and throughput requirements for large-scale processing ## References - [OpenAI API Pricing](https://openai.com/api/pricing/) - [OpenAI Prompt Caching](https://openai.com/index/api-prompt-caching/) - [Anthropic Claude Pricing](https://docs.claude.com/en/docs/about-claude/pricing) - [Google Gemini API Pricing](https://ai.google.dev/gemini-api/docs/pricing) - [Vertex AI Generative AI Pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing)