Thematic equity research focused on AI disruption
Infyro Research was founded in early 2026, with the mission of investigating the true impact of emerging artificial intelligence tools on software & Internet business models.
Loading articles...
Each company is scored 0–4 on five factors. Two measure risk (higher = more exposed); three measure protection or upside (higher = more shielded, or more of a beneficiary). The Net AI Disruption Score averages each side to a common scale, then takes the difference: avg(Pricing exposure, Margin exposure) − avg(Moat durability, Demand tailwind, AI response). Positive = net disruption risk. Negative = net AI beneficiary. Range runs roughly −4 to +4.
| Factor | What it measures | Direction |
|---|---|---|
| Seat/license pricing exposure | Share of revenue tied to per-seat or per-license pricing that AI agents could compress or let customers automate around | Risk (0–4) |
| Margin exposure to AI/inference costs | Whether embedding AI features pressures gross margin, versus the company being a supplier that benefits from others' AI capex | Risk (0–4) |
| Data & workflow moat durability | How defensible the incumbent's proprietary data, integrations, and switching costs are against an AI-native challenger | Protective (0–4) |
| AI-driven demand tailwind | Direct revenue/volume upside from other companies' AI buildout — chips, power, cooling, automation, logistics | Upside (0–4) |
| Competitive AI response | How credible and fast the company's own AI product strategy is at defending or growing share | Protective (0–4) |
Scores are a first-pass, directional read based on thesis groupings and company notes, not yet a fully researched pass in the way the Tokenomics figures are sourced and cross-checked. Treat them as a starting hypothesis per company, to be refined with real diligence (earnings-call language, disclosed AI-related revenue, customer commentary) before drawing firm conclusions.
| Provider | Model | Date | Input $/M | Output $/M | Note |
|---|
| Provider / Model | Tier | Input $/M tokens | Output $/M tokens |
|---|---|---|---|
| OpenAI | |||
| GPT-6 Astra | Flagship | $10.00 | $50.00 |
| GPT-5.6 Terra | Balanced | $2.00 | $12.00 |
| GPT-5.6 Luna | Fast / affordable | $0.20 | $1.20 |
| Anthropic | |||
| Claude Fable 5.1 | Flagship | $10.00 | $50.00 |
| Claude Opus 5 | High-capability | $5.00 | $25.00 |
| Claude Sonnet 5 | Balanced | $2.00 | $10.00 |
| Claude Haiku 4.5 | Fast / affordable | $1.00 | $5.00 |
| Gemini 3.1 Pro Preview | Flagship (≤200k / >200k ctx) | $2.00 / $4.00 | $12.00 / $18.00 |
| Gemini 3.8 Flash | Balanced (intro pricing to Dec 2026) | $0.75 | $3.75 |
| Gemini 3.1 Flash-Lite | Fast / affordable | $0.25 | $1.50 |
| xAI | |||
| Grok 4.6 | Flagship (≤200k / >200k ctx) | $2.00 / $4.00 | $6.00 / $12.00 |
| Grok 4.3 | Balanced | $1.25 | $2.50 |
| Meta | |||
| Llama 4 Maverick | Flagship | $0.19 | $0.65 |
| Llama 4 Scout | Balanced | $0.10 | $0.30 |
| Llama 3.3 70B | Fast / affordable | $0.10 | $0.32 |
| Mistral | |||
| Mistral Large 3 | Flagship | $0.50 | $1.50 |
| Mistral Medium 3.5 | Balanced | $1.50 | $7.50 |
| Mistral Small 4 | Fast / affordable | $0.15 | $0.60 |
| DeepSeek | |||
| DeepSeek V4 Pro | Flagship | $0.66 | $1.98 |
| DeepSeek Flash | Fast / affordable | $0.15 | $0.60 |
| Amazon | |||
| Nova Premier | Flagship | $2.50 | $12.50 |
| Nova Pro | Balanced | $0.80 | $3.20 |
| Nova Micro | Fast / affordable | $0.035 | $0.14 |
List API pricing as published by each provider (OpenAI, Anthropic, Google, xAI, Mistral, DeepSeek), correct as of September 2026. Meta and Amazon figures reflect third-party hosted pricing (via OpenRouter) rather than a simple official flat rate, since neither sells direct API access the same way. DeepSeek publishes separate peak-hour pricing (roughly 2x the off-peak rates shown). Excludes prompt-caching, batch, and long-context surcharge discounts, which vary by provider — see each provider's own pricing page for the fine print. Quality-adjusted value (price relative to benchmark/task performance) is a planned follow-on, not yet built.
Averages and cheapest/priciest above are computed across all 23 models in the table. The two context-tiered models (Gemini 3.1 Pro Preview, Grok 4.6) use their base (≤200k context) rate for this calculation.