CoachPal · cost tool

Live voice, priced honestly

What a month of AI coaching sessions costs across every current price tier of ChatGPT Realtime, Gemini Live (native and stitched pseudo-live), and Hawktalk — same minutes, same math, sourced rates. Every number is editable.

—

ChatGPT / mo

—

Gemini / mo

—

Hawktalk / mo

Backends — pick the tier to compare

ChatGPT Realtime

Gemini

Hawktalk

TRULY LIVE = native realtime audio, one stream, barge-in. PSEUDO-LIVE = stitched STT → text LLM → TTS pipeline; cheaper but adds turn latency and glue code the native options don't have. DEPRECATED = Google scheduled it for shutdown; shown for reference only.

Usage

Coaching seats (users)

Minutes of live conversation per session

Sessions per user per month

AI talk share — % of session the coach AI is speaking

Derived load

Total live minutes / month: —
User-speech minutes (audio input): —
AI-speech minutes (audio output): —
Sessions / month: —

Native-audio tiers: in_tok = user_min × in_tok/min, out_tok = ai_min × out_tok/min, cost = in/1M×$in + out/1M×$out. Hawktalk: (in+out)/1M × $1.00 flat. Pseudo-live: STT $/min × user_min + text-LLM tokens + TTS $/Mchar × chars. Token/char-per-minute rates are each provider's own published tokenization — all editable below.

Deliberately conservative in the incumbents' favor: only spoken audio is billed. Real realtime bills also re-bill conversation context each turn (cached input mitigates, not erases), which pushes native-tier bills higher than shown. Hawktalk's flat rate already includes everything.

Monthly cost

ChatGPT

Gemini

Hawktalk

$ per month, linear scale — hover a bar for the cost breakdown

Backend		Monthly	Per session	Per seat / mo	vs Hawktalk

Edit assumptions — rates & tokenization for the selected tiers

Nothing here is rigged: change any number and the whole page recalculates. Defaults are the providers' published rates, pulled 2026-07-17 (source link under each column). An EDITED tag appears on any tier you've changed.

Sources — all rates pulled 2026-07-17 from the providers' published pricing pages.
· OpenAI Realtime (developers.openai.com/api/docs/pricing): gpt-realtime-2.1 audio $32/M in · $0.40/M cached-in · $64/M out; gpt-realtime-2.1-mini $10 / $0.30 / $20. Audio token rates 1 tok/100 ms in, 1 tok/50 ms out (platform.openai.com/docs/guides/realtime-costs).
· Gemini Live API (ai.google.dev/gemini-api/docs/pricing): gemini-3.1-flash-live-preview audio $3/M in · $12/M out (Google also lists ≈$0.005/min in · $0.018/min out); gemini-2.5-flash-native-audio-preview-12-2025 audio $3/M in · $12/M out. Both are preview pricing. gemini-2.0-flash-live-001 ($0.35/M in · $1.50/M out, 25 tok/s) is third-party-sourced and was scheduled for deprecation 2025-12-09 — reference only. Audio tokenization ~32 tok/s in / 25 tok/s out per Google's token docs.
· Pseudo-live pipeline: Cloud Speech-to-Text standard recognition ≈$0.024/min at volume, Chirp realtime ≈$0.064/min (cloud.google.com/speech-to-text/pricing); Gemini 2.5 Flash-Lite text $0.10/M in · $0.40/M out (ai.google.dev pricing page); Cloud Text-to-Speech $4/M chars (Standard/WaveNet) up to $30/M chars (Chirp 3 HD) (cloud.google.com/text-to-speech/pricing). Speech-density defaults: ~150 words/min ⇒ 200 text tok/min and 800 chars/min.
· Hawktalk: $1.00/M tokens flat for live voice — list price; tokenization defaulted to the densest incumbent rate (1920/1500 tok/min) so Hawktalk is charged worst-case tokens.
· Model: input accrues while the user speaks, output while the AI speaks; silence, context re-billing, and pipeline glue are excluded, which understates the non-Hawktalk bills. No seat fees, infra, or minimums on any side.