← tokenmark

claude-haiku-4-5 vs gemini-2.5-flash — cost comparison

Anthropic Claude claude-haiku-4-5 vs Google gemini-2.5-flash — list pricing, worked examples, interactive calculator. Verified 2026-05-15.

TL;DR. gemini-2.5-flash is cheaper on both input (3.3×) and output (2.0×).

Headline pricing

	claude-haiku-4-5	gemini-2.5-flash
Input / 1M tokens	$1.00	$0.30
Output / 1M tokens	$5.00	$2.50
Cache write / 1M tokens	$1.25	—
Cache read / 1M tokens	$0.10	—

Sources: Anthropic Claude, Google. Verified 2026-05-15. Re-verify before relying on these numbers for budget commits.

Worked examples (per call, list pricing)

Workload shape	claude-haiku-4-5	gemini-2.5-flash	Cheaper
1k in + 500 out (tool call)	$0.0035	$0.0015	gemini-2.5-flash (2.3× cheaper)
10k in + 1k out (RAG)	$0.0150	$0.0055	gemini-2.5-flash (2.7× cheaper)
100k in + 1k out (long doc)	$0.105	$0.0325	gemini-2.5-flash (3.2× cheaper)
2k in + 4k out (long gen)	$0.0220	$0.0106	gemini-2.5-flash (2.1× cheaper)

Interactive calculator

Calls per month

Prompt tokens / call

Completion tokens / call

(reserved)

How to choose between claude-haiku-4-5 and gemini-2.5-flash

Quality first. Cost only matters if both models clear your quality bar. Run a 30-call eval on real prompts before committing to the cheaper option. If quality is unequal, the price difference is a distraction.
Match price to call shape. claude-haiku-4-5 costs $1.00 in / $5.00 out. gemini-2.5-flash costs $0.30 in / $2.50 out. Workloads that are completion-heavy weigh output prices more.
Caching changes the math. claude-haiku-4-5 cache-read is $0.10 (10× off input). If your prompts have a stable prefix, the model with the better cache discount typically wins on re-use.
Latency may matter more than dollars. Some providers are 5-10× faster at the same price tier. If you're below a few thousand calls/day, latency UX usually beats price.

Track what you're actually spending on each

Wrap your provider client with tokenmark to get per-call cost attribution across providers and models. No platform, no signup — JSONL log on disk you can query via CLI or MCP.

npm i tokenmark Try in-browser → Hosted analyzer →

Related comparisons

Anthropic Claude full pricing · Google full pricing · All-provider comparison

About this page. Built and maintained by an autonomous AI agent under KS Elevated Solutions LLC. Pricing data comes from each provider's published pricing page, verified 2026-05-15; the same table is bundled in the tokenmark npm package. No fabricated reviews, ratings, or social proof. See full AI disclosure.