💰 AI API costs for BYOK financial analysis
Complete 2026 guide — how much you really pay when using Alpha with your own Claude/OpenAI/Gemini key.
🤔 What is the BYOK model?
BYOK means Bring Your Own Key. Instead of paying a SaaS subscription several hundreds of euros per year that includes "AI", you pay your AI provider directly (Anthropic, OpenAI, Google, etc.) at cost. Alpha is just an orchestration layer: 46+ pre-configured analysis modules that call the API of your provider of choice with the right prompts.
You subscribe (€9.99/month), and AI usage is at market price, no middleman.
📊 Main provider pricing (2026)
| Provider | Flagship model | Input ($/M tokens) | Output ($/M tokens) | Recommended for |
|---|---|---|---|---|
| 🤖 Claude (Anthropic) | Opus 4.7 | $15 | $75 | 10-K, taxation, nuanced reasoning |
| 🤖 Claude Sonnet 4.6 | Balanced | $3 | $15 | Daily usage |
| 🤖 Claude Haiku 4.5 | Fast | $0.80 | $4 | Quick analyses, chatbot |
| 🧠 OpenAI GPT-5 | Flagship | $5 | $15 | Versatile |
| 🧠 GPT-5 Mini | Balanced | $1 | $4 | Standard analyses |
| 🧠 GPT-5 Nano | Fast | $0.20 | $0.80 | Cheapest OpenAI |
| ✨ Google Gemini | 2.5 Pro | $1.25 | $10 | Long context, native PDF |
| 🐦 xAI Grok | Grok 4 | $3 | $15 | Real-time X sentiment |
| 🇫🇷 Mistral | Large | $2 | $6 | EU data, code |
| ⚡ Cerebras | Llama 70B | $0.85 | $1.20 | Ultra-fast inference (>2000 tok/s) |
| 🐙 GitHub Models | Multi | Free* | Free* | Free tier (rate-limit) |
*Free tier with rate limits — personal use only with a GitHub PAT.
🎯 Estimated cost per Alpha module
Here are average costs per analysis for the most-used modules, based on real token profiles (input + output). Numbers shown use Claude Sonnet 4.6 (balanced tier), the app default.
| Module | Input/Output (tokens) | Cost/analysis | ×100 analyses |
|---|---|---|---|
| ⚡ Quick Analysis | 1500 / 800 | $0.017 | $1.70 |
| 💬 Chatbot (per message) | 1000 / 600 | $0.012 | $1.20 |
| 🎯 Position Sizing | 1200 / 800 | $0.015 | $1.50 |
| 🧮 DCF / Fair Value | 2500 / 2000 | $0.038 | $3.75 |
| 🌍 Macro Dashboard | 3000 / 2500 | $0.047 | $4.65 |
| 🇫🇷 Tax Optimizer FR | 3500 / 3500 | $0.063 | $6.30 |
| 🔥 Hidden fees (AI synthesis) | 4000 / 3500 | $0.065 | $6.50 |
| 📊 Portfolio Audit | 6000 / 4500 | $0.086 | $8.55 |
| 🚀 Research Agent | 8000 / 4000 | $0.084 | $8.40 |
| 📑 10-K Decoder | 25,000 / 5000 | $0.150 | $15.00 |
| 🎙 Earnings Call | 30,000 / 4000 | $0.150 | $15.00 |
| 🎙 YouTube + CEO Forensics | 35,000 / 4000 | $0.165 | $16.50 |
💡 Tip: If you mostly use light modules (Quick Analysis, Position Sizing, Chatbot), 100 analyses/month will cost you about $1.50. Less than a Starbucks coffee. Heavy analyses (10-K, Earnings, YouTube) should be reserved for important decisions — 5/month = $0.75. Total: ~$2-3/month for casual usage.
📈 Monthly budget by profile
🐢 Casual
5 analyses / week
Daily Quick Analysis + 1 DCF per week. Ideal for tracking 5-10 names.
⚡ Active
3 analyses / day
Daily Quick Analysis + DCF + Macro + Sentiment. Active investor tracking 30+ names.
🚀 Power User
Pro research
Daily Research Agent + 10-K + Earnings + YouTube. Amateur hedge-funder, financial advisor.
⚖️ BYOK vs SaaS comparison
| Solution | Model | Yearly price | AI included? |
|---|---|---|---|
| Alpha (BYOK) | Subscription + your API | ~€220/year total (€9.99/mo × 12 + ~$100 API) | ✅ You, at cost |
| Bloomberg Terminal | SaaS Pro | $24,000/year | ❌ No integrated generative AI |
| Koyfin Plus | Retail SaaS | $468/year | ❌ No AI |
| Simply Wall St | SaaS | $288/year | ⚠️ Limited proprietary AI |
| Stock Rover Premium+ | SaaS | $280/year | ❌ No AI |
| TradingView Premium | SaaS | $700/year | ❌ No generative AI |
💡 How to reduce API costs
- Default to "balanced" tier — Sonnet 4.6 or GPT-5 Mini are 5× cheaper than flagship for 90% of cases.
- Configure several providers and let the app's smart router pick the cheapest per module.
- Enable GitHub Models or Cerebras — free tier (rate-limited) covers casual usage.
- Disable web search on modules that don't need it (cache recent answers).
- Avoid duplicate analyses — the app keeps IndexedDB history, re-read it before re-running.
- Leverage prompt caching on Claude/OpenAI: if you chain analyses on the same asset, the system prompt is cached and billed 90% less.
🪙 Built-in optimizations in Alpha (automatic savings up to -70%)
Alpha includes 7 cost optimizations toggleable in one click in Settings → Advanced. All individually disableable.
| Optimization | How it works | Savings |
|---|---|---|
| 🪙 Eco Mode | Forces balanced tier (Sonnet 4.6, GPT-5 Mini, Gemini Flash) instead of flagship for all modules. | -70% |
| ⚡ 24h result cache | If you re-run a module with the same input within 24h, we reuse the existing IDB result. "Re-run" button available if you want to force. | -30 to -40% |
| 💾 Anthropic prompt caching | Marks system prompt as cacheable on Anthropic side (cache_control header). 2nd call billed 90% less on cached part. | -90% on cached tokens |
| 🎯 Wealth context trim | Wealth context sent to the AI is filtered per module: Tax-FR only gets FR holdings, IFI Simulator only gets real estate, etc. | -30 to -50% input tokens |
| 📡 15-min data context cache | FMP/Finnhub/CoinGecko calls (live prices, fundamentals) cached 15 min. 5 AAPL analyses in 5 min = 1 fetch instead of 5. | -50% on stocks API |
| 💰 Per-analysis budget cap | Set a $ cap — analysis is canceled if it would exceed the threshold. Anti-surprise guardrail. | Guardrail |
| 🆓 Free providers | Onboarding suggests GitHub Models / Cerebras / Mistral free tier. Covers ~50 analyses/month for free — how to get your API key for free in 5 minutes. | -100% casual |
🧮 Wealth-side savings on top
Beyond reducing API costs, Alpha includes 3 modules that save you directly on your wealth:
- 🧮 Tax-Loss Harvesting — detects latent losses on your CTO positions and proposes optimal sales to materialize losses (deductible 10 years in FR). Typical savings: €500-€2000/year on a €100K portfolio.
- 🔍 Subscription detector — scans your last 6 months of budget, spots duplicates (Netflix + Disney+ + Apple TV...). Typical savings: €20-50/month found on average.
- 🇫🇷 FR tax wrapper optimizer — recommends ideal PEA/AV/PER/CTO split based on age + marginal tax bracket. Avoids allocation mistakes that cost in taxes.
📉 Before / After in numbers
| Profile | No optims | With optims enabled | Savings |
|---|---|---|---|
| 🐢 Casual (5/week) | $8/mo | $2/mo | -75% |
| ⚡ Active (3/day) | $40/mo | $10/mo | -75% |
| 🚀 Power user | $150/mo | $45/mo | -70% |
Cumulative savings. Enable in 2 clicks via Settings → Advanced.
❓ FAQ
How much does an AI financial analysis cost?
Between $0.001 and $0.50 depending on the module. A quick analysis (Quick Analysis, ~2k tokens) costs ~$0.005. A deep 10-K report analysis (~30k input tokens) costs $0.10-$0.50 depending on the model used (Claude Opus, GPT-5, Gemini 2.5 Pro).
What's a realistic monthly budget to use Alpha in BYOK?
Depending on your usage: $2-$8/month for casual usage (5 analyses/week), $15-$40/month for active usage (3 analyses/day), $50-$150/month for power user (research-agent + 10-K + youtube daily). That's 100 to 200× cheaper than a Bloomberg Terminal.
What's the difference between fast / balanced / flagship tiers?
Fast: lightweight models (Claude Haiku, GPT-5 Nano) — $0.20-$0.80/M tokens — quick but less nuanced. Balanced: main models (Sonnet 4.6, GPT-5 Mini, Gemini 2.5 Pro) — $1-$3/M tokens — best price/quality ratio, recommended default. Flagship: top-tier models (Opus 4.7, GPT-5, Grok 4) — $5-$15/M tokens — for complex analyses (10-K, taxation, long reasoning).
How to reduce API costs?
1) Use the "fast" tier for simple analyses. 2) Configure several providers and let the smart router pick the cheapest per task. 3) Enable Cerebras or GitHub Models (free tier with rate-limit). 4) Disable web search for modules that don't need it. 5) Leverage Anthropic prompt caching (90% off on repeated prompts).
Why is BYOK cheaper than a SaaS subscription?
An AI analysis SaaS takes 5 to 10× the raw API cost to cover its cloud infra, customer support, commercial margin. With BYOK you pay the API at cost from the provider (Claude, OpenAI, Gemini). Typical savings: a user at 50 analyses/month pays $5-$15 in BYOK vs $30-$50 in equivalent SaaS subscription. And if you don't use the app for 1 month, you pay nothing.
Are API keys stored securely?
Yes. Keys are encrypted locally on your device via AES-GCM 256 + PBKDF2 100K iterations. The master password never leaves your machine. Alpha has no servers, no proxy, no telemetry. You can verify with DevTools → Network: only direct calls to provider APIs are visible.
🚀 Ready to try?
€9.99/month, your API at cost. Free demo mode to test without configuring a key.
Launch Alpha →