Complex-doc accuracy
99.9%
Positioned for complex layouts, handwriting, and harder document classes.
Model lineup
Highest-scoring OCR models on Omnidoc benchmark, tuned for production
LeapOCR combines benchmark-leading model families with structure handling, QA, and refinement so you can run fast production OCR or push for maximum confidence on harder documents without changing your workflow.
Why this matters
You are not picking between a “basic” model and a “real” one.
Standard-v2 and pro-v2 are both production-grade OCR models. Standard-v2
is built for strong throughput across the common path. Pro-v2 is built for
the hardest pages and the strictest review standards.
Active engine status
Both model lanes are production-ready. Choose the one that matches your document mix, review tolerance, and speed requirements.
Omnidoc proven
Production lane
standard-v2
Fast, benchmark-backed OCR for invoices, forms, receipts, IDs, and most recurring document traffic.
Speed, range, and production volume
Base
1
Use standard-v2 when you want benchmark-backed quality at scale. Use pro-v2 when the queue is harder by default or the cost of misses is higher.
Choose the lane that fits
Base
3
Flagship lane
pro-v2
Highest-confidence extraction for handwriting, denser tables, noisy scans, and lower-tolerance workflows.
Harder docs and stricter review environments
Fast-lane latency
142ms
Reference latency callout used on the old models page for the faster production lane.
Production models
2
Standard-v2 and pro-v2 are both built for production, with different confidence and cost profiles.
Benchmark standing
Omnidoc leaders
LeapOCR is positioned around highest-scoring model families on Omnidoc benchmark.
Precision orchestration
Higher accuracy comes from the full extraction chain
Stage 01
Vision OCR
Layout, text, and structure detection start the extraction chain.
Stage 02
LLM QA
Vision-aware QA checks and corrects output before it leaves the pipeline.
Stage 03
Refinement
Normalization, schema shaping, customization, and bbox sit on top when needed.
Powered by
Gemini
Qwen
Open-source VLMs
Custom processing
LeapOCR layers structure handling, QA, normalization, and output shaping on top of the model itself, which is why the result is stronger than raw OCR alone.
Side By Side
Choose by operating posture
standard-v2
Benchmark-backed production model for speed, structure, and range
Most deployed
Standard-v2 is not a fallback tier. It is a strong production OCR model built on top-ranked document engines, then improved through LeapOCR's structure, QA, and refinement pipeline.
Positioning
Best balance of throughput, confidence, and cost efficiency
Base credits
1 base credit / page
Best for
- Invoices, receipts, forms, IDs, and structured business paperwork at production volume
- Teams that want fast, benchmark-proven OCR without giving up layout awareness
- Pipelines where standardization and clean output matter as much as raw text capture
Tradeoffs
- Can still benefit from refinement or pro-v2 on the worst scans and hardest layouts
- Not always the ideal choice for the lowest-tolerance edge-case queues
pro-v2
Flagship accuracy lane for the hardest pages and strictest workflows
Highest confidence
Pro-v2 is the highest-confidence model in the lineup, built for difficult pages, layout noise, handwriting, denser tables, and workflows where the cost of a miss is materially higher.
Positioning
Best for maximum extraction confidence on harder documents
Base credits
3 base credits / page
Best for
- Messy scans, irregular layouts, handwriting, denser tables, and visually noisy pages
- Compliance, finance, and operations workflows with tighter review tolerance
- Teams that want the strongest model from the first pass on higher-value queues
Tradeoffs
- Higher base credit cost than standard-v2
- Usually best reserved for harder queues rather than every low-risk page
Decision guide
Pick the stronger fit for the queue you actually run
Both models are worth deploying. Standard-v2 wins when you need strong quality at scale. Pro-v2 wins when your documents are consistently harder or when the error budget is tighter from the start.
Practical rollout
Start with the model that matches your queue, not a rule of thumb.
If most of your workload is business paperwork, standard-v2 is usually the right production lane. If most of it is messy, irregular, or high-stakes, pro-v2 is a valid first choice, not just an escalation target.
Bulk business paperwork
For invoices, receipts, purchase orders, common forms, and most business paperwork, standard-v2 already delivers a strong production baseline.
standard-v2
Messy or irregular documents
When the queue shifts toward handwriting, layout chaos, denser tables, or harder scans, pro-v2 is the higher-confidence lane.
pro-v2
Mixed production workloads
Many teams keep both models live: standard-v2 for the main queue and pro-v2 for premium queues, failed checks, or the hardest document classes.
Route by queue
Ready to test
Run your own documents through both models and compare the result.
FAQ
Common model questions
Are these benchmark-backed models or custom OCR from scratch?
LeapOCR is positioned around highest-scoring model families on Omnidoc benchmark, then improves them through a custom pipeline for structure handling, QA, refinement, and output shaping.
Is standard-v2 already good enough for production?
Yes. Standard-v2 is a serious production model, not a stripped-down teaser. It is designed for most recurring business-document queues and performs strongly on the common path.
Why would I choose pro-v2 from the start?
Use pro-v2 from the first pass when the queue is harder by default or when extraction mistakes are expensive enough that you want the highest-confidence lane immediately.
How do model credits work?
Standard-v2 starts at 1 base credit per page. Pro-v2 starts at 3 base credits per page. Refinement and bbox layers add on top when enabled.