Engine Pricing
Important Note: We've introduced Smart Context Compression (formerly Palo Output). This features intelligently abstracts and compresses your input context, significantly reducing your effective token usage. However, for 770 and DEEP-R engines, this minimization will not apply to code or research data, which will be processed fully by the external LLM to maintain precision.
Subscription Plans: Architect ($20/month) includes 120M tokens. Business ($35/user/month) includes 200M tokens. Overage is billed at the rates below. See Billing for full plan details.
| Engine / API Name | Mini (Lite) (mpalo-palo-lite) | Palo Bloom (mpalo-palo) | 770 (mpalo-palo-770) | DEEP-R (mpalo-palo-DEEP-R) |
|---|---|---|---|---|
| Blended In/Out Rate (per 1M tokens) | $0.3 | $0.9 | $2.1 | $2.9 |
Memory Traversal
Optional: Traversal of memories for better supported recall, Episodic.
Priced at 60% of blended I/O rate.
(optional, per 1M tokens)
|
$0.18 | $0.54 | $1.26 | $1.76 |
| Memory Mapping
Optional: Mapping of memories for more accurate supported recall,
Episodic + Temporal. Priced at 80% of blended In/Out rate.
(optional, per 1M tokens)
|
$0.24 | $0.72 | $1.68 | $2.25 |
| Image Processing (per image) | $0.0005 | $0.001 | $0.003 | $0.015 |
| Audio Processing (Q4 2026 Beta) |
~$1.50 in / $6 out | ~$2.00 in / $8 out | ~$3.00 in / $10 out | ~$5.00 in / $12 out |
The Blended In/Out Rate combines input and output pricing for simplicity. Memory Traversal and Memory Mapping are optional features charged only when used.
How it works: All intermediate token costs are calculated as 60% (Traversal) and 80% (Mapping) of the engine's blended In/Out rate, ensuring predictable and proportional costs regardless of your workflow. This means if you use more memory features, costs scale fairly with the engine's power level.
Show detailed pricing breakdown for reference
| Engine / API Name | Mini (Lite) (mpalo-palo-lite) | Palo Bloom (mpalo-palo) | 770 (mpalo-palo-770) | DEEP-R (mpalo-palo-DEEP-R) |
|---|---|---|---|---|
| Input Price (per 1M tokens) | $0.235 | $0.806 | $1.8 | $2.484 |
| Output Price (per 1M tokens) | $0.355 | $1.032 | $2.34 | $3.229 |
| Memory Traversal (80% of blended, per 1M tokens) | $0.18 | $0.54 | $1.26 | $1.76 |
| Memory Mapping (80% of blended, per 1M tokens) | $0.24 | $0.76 | $1.68 | $2.25 |
| Image Processing (per image) | $0.0005 | $0.001 | $0.003 | $0.015 |
| Audio Processing (Est. per 1M) | Betas Q2 '26 | Betas Q2 '26 | Betas Q2 '26 | Betas Q2 '26 |
Note: Prices shown are illustrative for Mpalo Palo engine usage only.
Storage Costs (BYOVS): You choose your vector storage provider (e.g., Pinecone,
Weaviate). Storage costs are paid directly to them (typically $0.10–$0.50/GB-month). With Mpalo,
you pay for the intelligent memory access (Traversal/Mapping) to utilize that data, not
the passive storage itself.
Billing for external AI models (OpenAI, Google) is also separate and handled via your API
keys.
Feature Decision Guide: Traversal vs. Mapping
| Feature | Best For | Recall Type | Performance |
|---|---|---|---|
| Memory Traversal | Chatbots, FAQs, simple personalization | Episodic (Event-based) | Fast (~50ms latency), Lower Cost |
| Memory Mapping | Complex Agents, Long-term reasoning, Research | Episodic + Temporal (Time-aware) | Deeper (~200ms latency), Higher Accuracy |
Add-ons & Extras
| Add-on | Price | Availability |
|---|---|---|
| Custom Connections (Ollama, etc.) | $5 / slot / month | Free on Architect (2 slots) & Business (5 slots) |
| Secure Tunnel | $20 / month | Business & Enterprise Plans |
| Private Data Spaces | Usage-based | Enterprise Only |
Monthly Cost Calculator
To provide a helpful starting point, Memory Traversal and Mapping costs are automatically estimated based on your input size (4.5x and 5.3x respectively). The total is calculated using the detailed pricing table above.
Estimated Monthly Cost:
Feature Limitations
| Feature | Mini | Palo Bloom | DEEP | DEEP-Research |
|---|---|---|---|---|
| Max Image Size | 5 MB | 10 MB | 50 MB | 100 MB |
| Image Formats | JPG, PNG | JPG, PNG, GIF | JPG, PNG, GIF, WebP, TIFF, BMP | All formats + RAW |
| Max Images per Call | 1 | 3 | 10 | 25 |
| Max Audio Size (TBD) | 5 MB | 10 MB | 50 MB | 100 MB |
| Audio Formats (TBD) | MP3, WAV | MP3, WAV, AAC | MP3, WAV, AAC, FLAC, OGG | All formats + hi-res |
| Max Audio per Call (TBD) | 1 | 3 | 10 | 25 |
Understanding Mpalo's Capabilities and Limitations
It is crucial to understand that Mpalo's Palo Engines do not generate images or audio based on prompts. Instead, our models are designed to summarize, memorize, and process visual and auditory input to build rich, persistent memory representations.
Image Input Requirements
Input images must meet the following requirements to be processed by the API:
- Supported File Types: PNG, JPEG, WEBP, and non-animated GIF.
- Size Limits: Up to 20MB per image.
- Quality: Images are resized before analysis, which may affect original dimensions.
- Content: No watermarks, logos, or NSFW content. Must be clear enough for a human to understand.
Additional Model Limitations
While our vision capabilities are powerful, it's important to understand their current limitations:
- Non-English Text: May not perform optimally with non-Latin alphabets.
- Small Text: Enlarge text within the image for better readability.
- Rotation: May misinterpret rotated or upside-down content.
- Visual Elements: May struggle with graphs or text where meaning relies on colors or styles.
- Spatial Reasoning: Struggles with tasks requiring precise spatial localization.
- Counting: May provide approximate counts of objects.
We are actively working to enhance these capabilities.