llm api pricing
Per-1M-token list prices
| Model | Vendor | Input / 1M | Output / 1M | Cached input |
|---|---|---|---|---|
| Claude Haiku 3.5 | Anthropic | $0.80 | $4.00 | $0.08 input |
| Claude Sonnet 4.6 | Anthropic | $3.00 | $15.00 | $0.30 input |
| Claude Opus 4 | Anthropic | $15.00 | $75.00 | $1.50 input |
| GPT-4o-mini | OpenAI | $0.15 | $0.60 | $0.075 input |
| GPT-4o | OpenAI | $2.50 | $10.00 | $1.25 input |
| GPT-5 | OpenAI | $8.00 | $32.00 | $4.00 input |
| Gemini 2.0 Flash | $0.10 | $0.40 | $0.025 input | |
| Gemini 2.0 Pro | $1.25 | $5.00 | $0.31 input |
Monthly cost at 4 common SME workloads
Picking a model in the abstract is hard; picking one against a specific workload is easy. Below: 4 NodeSparks reference architectures and what each costs across 5 models.
| Use case | Haiku | Sonnet | GPT-4o-mini | GPT-4o | Gemini Flash |
|---|---|---|---|---|---|
| Small outreach agent (500 emails/mo personalized)500k in / 50k out | $0.60 | $2.25 | $0.11 | $1.75 | $0.07 |
| Slack invoice agent (200 invoices/mo with vision)4M in / 200k out | $4.00 | $15.00 | $0.72 | $12.00 | $0.48 |
| Mid-size ops automation (10k workflow runs/mo)50M in / 5M out | $60 | $225 | $10.50 | $175 | $7 |
| Large content engagement agent (1M social actions/mo)500M in / 50M out | $600 | $2,250 | $105 | $1,750 | $70 |
Per-model notes
- Claude Haiku 3.5. Fastest Claude. Best for classification, extraction, summarization at scale. 10× prompt-cache discount on cached tokens.
- Claude Sonnet 4.6. Default workhorse. Best balance of cost, reasoning, and speed for most production use. 200k context window.
- Claude Opus 4. Highest reasoning capability. Use only when Sonnet visibly struggles. Premium price for premium tasks.
- GPT-4o-mini. Cheapest competent model. Best for high-volume personalization, simple classification. 128k context.
- GPT-4o. OpenAI workhorse. Comparable to Claude Sonnet on most tasks; cheaper. 128k context.
- GPT-5. OpenAI flagship reasoning model. Use sparingly — most tasks do not need GPT-5 quality.
- Gemini 2.0 Flash. Cheapest in the comparison. 1M context window. Best for ingesting large documents (RAG, code review at scale).
- Gemini 2.0 Pro. Comparable to Claude Sonnet on cost; weaker on reasoning, stronger on multimodal. 2M context window.
Methodology
List prices captured on 2026-06-02 from each vendor's public pricing page (anthropic.com/pricing, openai.com/api/pricing, ai.google.dev/pricing). Volumes assume single-region US/EU endpoint deployment; cross-region inference is typically ~10% more expensive.
Workload examples assume average prompt size and a reasonable cache hit rate (~60% for the Sonnet/GPT-4o numbers). Real-world cost can deviate ±30% depending on prompt engineering quality.
How to cite
NodeSparks (2026). "LLM API Pricing 2026 — Real Cost at SME Volumes." https://www.nodesparks.com/data/llm-api-pricing. CC BY 4.0.
Related
Download
⬇ Download CSV
Per-1M-token list prices, CC BY 4.0.
Last updated 2026-06-02.
Still paying for tools you could own?
We replace the SaaS stack and the manual ops work eating your team's time. One custom system, owned by you.
We replace the SaaS stack and the manual ops work eating your team's time. One custom system, owned by you.