The moment you need a model that can juggle nuanced reasoning and fast turn‑around, the per‑token bill becomes the litmus test – Claude Sonnet 5 asks $2 for every million input tokens while Gemini 3 Flash‑lite charges $0.25 for the same work. That raw contrast alone can swing a budget‑conscious team toward Google, but the story deepens once you factor in output rates and cached‑input discounts.
### Where the dollars actually go
Claude Sonnet 5’s pricing sheet lists three line items: input $2, cached input $0.2, and output $10 per 1 M tokens. Google’s Gemini 3 Flash‑lite mirrors that structure with input $0.25, cached input $0.025, and output $1.5 per 1 M tokens. The cached‑input discount matters for workloads that reuse system prompts or few‑shot examples – Anthropic’s cache is eight times pricier than Google’s, turning a repeated‑prompt scenario into a hidden cost driver. Even if your prompts are fresh each call, the output differential is stark: Anthropic bills $10 for every million generated tokens versus Google’s $1.5, a six‑fold gap that can dwarf any input savings when the model produces long answers.
### Who each model serves best
If your application is a high‑throughput chatbot that spews out several hundred tokens per turn, Gemini 3 Flash‑lite’s low output fee makes it the clear economic winner. The model still offers respectable reasoning for a mid‑tier offering, and the cheap cached‑input rate keeps recurring system prompts affordable. On the other hand, Claude Sonnet 5 shines when you need the extra horsepower of Anthropic’s safety‑tuned instruction set and are willing to pay a premium for the higher output cost. Enterprises that run fewer, longer completions – for example, drafting legal clauses or detailed reports – may justify the $10 output rate because the total token count stays modest.
| Model | Input | Cached Input | Output |
|---|---|---|---|
| Claude Sonnet 5 (Anthropic) | $2 | $0.2 | $10 |
| Gemini 3 Flash‑lite (Google) | $0.25 | $0.025 | $1.5 |
In practice, the choice narrows to a trade‑off between raw cost and the nuanced safety and instruction handling Anthropic touts. For startups and developers racing to ship a feature‑rich chat experience, Gemini 3 Flash‑lite delivers the cheapest per‑token bill by a wide margin, keeping operational spend predictable. Large organisations that value Anthropic’s guardrails and are prepared to absorb higher output fees will find Claude Sonnet 5 a justifiable premium for the added assurance.
Pick Gemini 3 Flash‑lite if you need volume, low‑cost output, and can work within Google’s broader ecosystem. Opt for Claude Sonnet 5 when safety, instruction fidelity, and Anthropic’s model behavior outweigh the steep per‑token price. Pricing verified on 2026-09-12.