Three Clear Pricing Brackets Emerged Across Providers
Following our manual November 4, 2026 review of official provider pricing documentation across OpenAI, Anthropic, Google, DeepSeek, xAI, Cohere, and Mistral, active production API models cluster tightly into three cost brackets: flagship deep-reasoning tiers ($4–$10 input / $20–$50 output per 1M tokens), workhorse coding/production tiers ($1.25–$2.00 input / $2.50–$12.00 output), and high-throughput sub-dollar tiers ($0.10–$0.30 input).
For engineering teams building cost calculators or multi-tier routers, that 100x spread between $10.00/1M and $0.10/1M input pricing means model assignment per endpoint now moves cloud spend far more than minor prompt tweaks alone.
| Model | Provider | Input / 1M | Output / 1M | Cache Read / 1M |
|---|---|---|---|---|
| GPT-6 Astra | OpenAI | $10.00 | $50.00 | $1.00 |
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | $0.25 |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | $0.40 |
| Gemini 4 Argon | $4.00 | $20.00 | $0.40 | |
| GPT-6.1 Sol | OpenAI | $2.00 | $10.00 | $0.20 |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | $0.20 |
| DeepSeek V4 Pro | DeepSeek | $0.66 | $1.98 | $0.07 |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | $0.03 | |
| DeepSeek Flash | DeepSeek | $0.15 | $0.60 | $0.015 |
| GPT-6 Luna / Claude Haiku 5.5 | OpenAI / Anthropic | $0.10 | $0.50 | $0.01 |
Active tier snapshot from TokenCalc AI catalog (reviewed Nov 4, 2026)
Cache Read Pricing Favors Heavy Static Contexts
Alongside base token rates, cached input reads continue pushing repeated system prompts and code context costs down by roughly 90% or more relative to uncached input rates on supported models. Notably, Claude Fable 5.1 lists cache reads at $0.25/1M compared to GPT-6 Astra at $1.00/1M, altering break-even math for long-running agent threads.
- GPT-6 Luna and Claude Haiku 5.5 both price cache reads at $0.01/1M.
- DeepSeek Flash reaches $0.015/1M on cached prefix hits.
- Asynchronous Batch API discounts remain 50% off standard input/output rates on OpenAI, Anthropic, and Google tiers.
Scope & Verification Transparency
All numbers cited above reflect our static maintainer verification audit recorded on November 4, 2026 rather than automated runtime scraping. Where official pages omit hardware chips or Artificial Analysis throughput medians, fields remain explicitly unmarked in the verification hub.