AI Model Pricing Tracker
Every frontier model, every provider, sortable by anything. The reference table we wished existed when we were comparing bills.
| Cached /1M | Modalities | ||||||
|---|---|---|---|---|---|---|---|
| Gemini 3 Flash LiteFree tier | $0.10 | $0.40 | $0.03 | 1M | TextVision | 2026-01 | |
| Mistral Small 3Free tier | Mistral | $0.20 | $0.60 | — | 128K | Text | 2026-01 |
| DeepSeek V4Free tier | DeepSeek | $0.27 | $1.10 | $0.07 | 128K | Text | 2026-02 |
| Grok 4 MiniFree tier | xAI | $0.30 | $0.50 | — | 131K | Text | 2026-03 |
| Gemini 3 FlashFree tier | $0.35 | $1.40 | $0.09 | 1M | TextVisionAudio inVideo | 2026-01 | |
| DeepSeek R2 (reasoning) | DeepSeek | $0.55 | $2.19 | $0.14 | 64K | Text | 2026-04 |
| Claude 4.5 HaikuFree tier | Anthropic | $0.80 | $4.00 | $0.08 | 200K | TextVision | 2025-10 |
| o3 Mini (reasoning) | OpenAI | $1.10 | $4.40 | $0.55 | 200K | Text | 2025-01 |
| GPT-5 MiniFree tier | OpenAI | $1.50 | $6.00 | $0.15 | 200K | TextVision | 2025-12 |
| Qwen 3 MaxFree tier | Alibaba | $1.60 | $6.40 | $0.16 | 256K | TextVision | 2026-02 |
| GPT-4.1 | OpenAI | $2.00 | $8.00 | $0.50 | 1M | TextVision | 2025-04 |
| Mistral Large 3 | Mistral | $2.00 | $6.00 | — | 256K | Text | 2025-11 |
| Claude 4.6 SonnetFree tier | Anthropic | $3.00 | $15 | $0.30 | 1M | TextVision | 2026-02 |
| Gemini 3 ProFree tier | $3.50 | $14 | $0.88 | 2M | TextVisionAudio inVideo | 2026-01 | |
| Llama 4 405B (Together) | Meta | $3.50 | $3.50 | — | 128K | Text | 2025-09 |
| Grok 4Free tier | xAI | $5.00 | $15 | — | 256K | TextVision | 2026-03 |
| GPT-5Free tier | OpenAI | $13 | $50 | $1.25 | 400K | TextVisionAudio inAudio out | 2025-12 |
| Claude 4.7 Opus | Anthropic | $15 | $75 | $1.50 | 1M | TextVision | 2026-04 |
Prices are per 1 million tokens, as published by each provider. PKR conversion at 1 USD = 279 PKR. Some providers offer batch APIs, committed-use discounts, or off-peak pricing that materially reduces effective cost; this tracker reflects the standard published rate only.
How we keep this current
We check each provider's official pricing page weekly and refresh this table within 24 hours of any change. The data lives in a single JSON-shaped file in our repository, which means every update is version-controlled and visible in our public commit history.
If you spot a stale figure or missing model, please email faizan@cubitrek.comand we'll update within hours.
What this tracker covers (and doesn't)
- Covers: Standard published API pricing for text-out, including cached input where available. Context window size. Modality support. Free-tier availability.
- Doesn't cover: Batch API discounts (usually 50 percent off), committed-use contracts, regional discounts, promotional credits. These can materially change the real cost of a workload but vary too much by account to track centrally.
Frequently asked questions
How often is this tracker updated?
We refresh the data whenever a provider changes their published pricing. Major providers (Anthropic, OpenAI, Google) tend to update quarterly; smaller players move more often. The last-updated date at the top of the page reflects the most recent verified change.
What does input vs output pricing mean?
AI providers charge separately for tokens you send (input) and tokens the model returns (output). Output is typically 3 to 5 times more expensive than input. A typical chat workload has 5 to 10 times more input tokens than output, so input pricing usually dominates the bill.
What is cached input pricing?
Most major providers offer a discount when you re-send a system prompt or document the model has seen recently. Cached input pricing applies to that re-used portion, typically at 10 percent of the standard input rate. For RAG and agent workloads, cached pricing can cut bills by 40 to 70 percent.
Which model is actually cheapest?
It depends on the workload. For pure text at scale, DeepSeek V4 and Gemini 3 Flash Lite are the cheapest serious models on the market. For reasoning, DeepSeek R2 and o3 Mini are the lowest-cost. For vision, Gemini 3 Flash beats most competitors on $/MP. Use our cost calculator to estimate your specific workload.
Why are the prices in PKR rounded?
We convert at a fixed reference USD/PKR rate and round to whole rupees for readability. Your real cost will track the prevailing interbank rate at the time your card is charged, which can move 1 to 3 percent in either direction.
Related on Meridian48
The week in AI, Pakistan tech, and global business.
One email, every Friday morning. Curated and written by Faizan Khan. No filler. No tracking pixels. Unsubscribe in one click.