benchr
API pricing for large language models has commoditized rapidly, but the pricing structures have become more complex. Between input, output, prompt caching, batch execution, and self-hosted instances, developers must calculate pricing carefully to estimate their monthly workloads. Below is our dynamic pricing comparison table, kept in exact sync with the main model index.
| Model | Provider | Input / 1M | Output / 1M | Cached Input / 1M |
|---|---|---|---|---|
| Qwen3.6-27B | Alibaba | Self-hosted | Self-hosted | — |
| Llama 4 Maverick | Meta | Self-hosted | Self-hosted | — |
| Llama 4 Scout | Meta | Self-hosted | Self-hosted | — |
| Phi-4 | Microsoft | Self-hosted | Self-hosted | — |
| GPT-5 Mini | OpenAI | $0.250 | $2.00 | $0.025 |
| DeepSeek V4-Flash | DeepSeek | $0.440 | $1.32 | $0.014 |
| Mistral Large 3 | Mistral | $0.500 | $1.50 | — |
| Kimi K2.6 | Moonshot AI | $0.950 | $4.00 | $0.160 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | $0.100 |
| GPT-5 | OpenAI | $1.25 | $10.00 | — |
| Grok 4.3 | xAI | $1.25 | $2.50 | $0.200 |
| DeepSeek V4-Pro | DeepSeek | $1.32 | $3.96 | $0.044 |
| Gemini 3.5 Flash | $1.50 | $9.00 | $0.150 | |
| Mistral Medium 3.5 | Mistral | $1.50 | $7.50 | — |
| Gemini 3.1 Pro | $2.00 | $12.00 | $0.200 | |
| Claude Sonnet 4.6 | Anthropic | $3.00 | $15.00 | $0.300 |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 | $0.500 |
| Claude Opus 4.7 | Anthropic | $5.00 | $25.00 | $0.500 |
| GPT-5.5 | OpenAI | $5.00 | $30.00 | $0.500 |
DeepSeek is the only provider here whose rate depends on when the request runs. The figures shown are its peak rate; the weekday window 01:00–04:00 and 06:00–10:00 UTC is expensive and everything else is half price. What that means in practice.
Pricing Tiers: Frontier vs. Mid-Tier vs. Small
When analyzing costs, models generally fall into three tiers:
- Frontier Tier ($5.00+ Input / $15.00+ Output): This highest-cost group includes Claude Opus 4.8 and GPT-5.5. Test it when the expected quality gain can justify the premium; routine high-volume work often needs a cheaper route.
- Mid-Tier ($1.25 - $3.00 Input / $2.50 - $15.00 Output): Claude Sonnet 4.6, GPT-5, and Gemini 3.5 Flash sit here. Compare them when you need more capability than a budget tier without paying the highest listed rates.
- Small & Open Coder Tier (Sub-$1.00 Input): This group includes DeepSeek V4-Flash, GPT-5 Mini, and self-hosted Llama 4 Scout. Their lower listed rates make them candidates for routing, classification, and high-volume summarization, subject to task testing.
For more details on overall performance rankings, see our AI Model Rankings or compute your exact monthly cost with the Cost Calculator.
How to avoid false precision
Per-million-token pricing looks exact, but real invoices depend on workload shape. A support bot with a long policy prompt and short answers is mostly input cost; a writing assistant with short prompts and long drafts is mostly output cost. The same model can be cheap in one workflow and expensive in another.
Use the table as a normalized baseline, then calculate your own mix: average input tokens, average output tokens, cache hit rate, batch percentage, retry rate, and accepted-output rate. Those six numbers matter more than a one-line provider price when you are deciding what to ship.
Re-check the provider page before committing annual spend. Pricing pages change quietly, and some enterprise contracts include discounts, minimums, or platform fees that do not appear in public documentation.
The decision rule
Start with the least expensive model that clears a versioned task set, not the cheapest row in isolation. Run the same prompts and acceptance checks across a budget model and one stronger control, then price successful outputs rather than raw calls. Move up a tier only when the additional accepted work is worth more than the token premium; move down when retries, fallbacks, and review time remain inside your operating limit.