DEV Community

Manohar Papasani
Manohar Papasani

Posted on

Gemini 1.5 Pro's per-token price doubled past 128K tokens, so check the cost before you scale.

#191 #ai #genai #llm

Its input price was half of GPT-4o's, but BenchLM's quality score for it is only an estimate.

Sources: BenchLM, Activepieces and Puter pricing breakdowns.
Quality evidence is thin, so this post covers cost and limits.

  • What does a long prompt really cost?
  • Is there independent evidence on quality?

📌 THE 128K PRICE JUMP
Gemini 1.5 Pro (Google)

  • Cost per token doubled once a prompt exceeded 128,000 tokens (Activepieces).
  • Its $1.25 input price was the figure cited for it, labeled Previous in BenchLM's pricing table.

📌 PRICE VS RIVALS
GPT-4o (OpenAI)

  • $2.50 per 1M input tokens, double Gemini 1.5 Pro's input price (Activepieces). Claude 3.5 Sonnet (Anthropic)
  • $3.00 per 1M tokens; Gemini 1.5 Pro was 58% cheaper (Activepieces, citing Compute Comparison).

📌 WHAT THE QUALITY DATA SAYS
BenchLM score (BenchLM.ai)

  • Gemini 1.5 Pro scores 27.16/100 as an estimate, with public rank #191.
  • It shares 0 benchmark results with its comparison model, so BenchLM gives no quality verdict.

What this means for your team:

  1. Model your bill at your longest prompts, since past 128K tokens the per-token rate doubles.
  2. Compare price against GPT-4o and Claude 3.5 Sonnet, but don't assume quality from price alone.
  3. Ask for independent test results before you commit a production workload.
  4. Check current Google pricing, since newer Gemini models list separate rates.

Pull your largest prompt sizes this week and check them against the 128K line before you choose a model.

P.S. Newer Gemini 3.1 Pro lists $2.00 input and $4.00 above 200K tokens, so the long-prompt jump is still there.

Originally shared on LinkedIn.

An AI interview question every day, leadership for tech people and hands-on AI learning:

Top comments (0)