Gemini 3.8 Flash vs GPT-6 Luna
Same documents, same prompt: where each model is right, what it really costs and how fast it answers.
At a glance
| Task · documents | English (US) | Czech | German | Spanish | French |
|---|---|---|---|---|---|
| Invoice extraction | 99.8% / 99.7% | 100.0% / 99.3% | 97.2% / 99.2% | 95.9% / 99.1% | 99.9% / 98.9% |
| Email triage | 98.7% / 97.5% | 99.2% / 97.9% | 98.4% / 97.0% | — | — |
| Personal data detection | 99.7% / 99.6% | 99.5% / 99.6% | 99.7% / 99.6% | — | — |
| Contract clauses | 99.9% / 99.8% | 100.0% / 100.0% | 99.9% / 100.0% | — | — |
Correct fields: Gemini 3.8 Flash / GPT-6 Luna; the better one in bold. Click a cell to see the details.
Invoice extraction · English (US) (150 documents)
Gemini 3.8 Flash is more accurate by 0.1 pp. GPT-6 Luna is 7.6× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.8% | 99.7% |
| Error-free documents | 98% | 97% |
| Per 1,000 documents | $3.84 | $0.51 |
| To fix / 1,000 | 20 | 33 |
| Speed (median) | 3.2 sec | 3.4 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Supplier | 100.0% | 100.0% |
| Supplier ID (EIN, IČO…) | 100.0% | 100.0% |
| Invoice number | 100.0% | 99.3% |
| Payment reference | 100.0% | 98.0% |
| Invoice date | 100.0% | 100.0% |
| Due date | 100.0% | 100.0% |
| Net by tax rate | 100.0% | 100.0% |
| Tax by rate | 100.0% | 100.0% |
| Total | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Bank account | 98.0% | 99.3% |
Documents where they differ
- Raleigh Engineering Inc. · INV253511 · Scan · Gemini 3.8 Flash 11/11, GPT-6 Luna 10/11
- Rose Lowery · INV-14797 · Clean image · Gemini 3.8 Flash 10/11, GPT-6 Luna 11/11
Invoice extraction · Czech (150 documents)
Gemini 3.8 Flash is more accurate by 0.7 pp. GPT-6 Luna is 8.4× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.0% | 99.3% |
| Error-free documents | 99% | 93% |
| Per 1,000 documents | $4.26 | $0.51 |
| To fix / 1,000 | 7 | 67 |
| Speed (median) | 3.5 sec | 3.2 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Supplier | 99.3% | 95.3% |
| Supplier ID (EIN, IČO…) | 100.0% | 98.7% |
| VAT ID | 100.0% | 98.0% |
| Invoice number | 100.0% | 100.0% |
| Payment reference | 100.0% | 100.0% |
| Invoice date | 100.0% | 100.0% |
| Tax point | 100.0% | 100.0% |
| Due date | 100.0% | 99.3% |
| Net by tax rate | 100.0% | 100.0% |
| Tax by rate | 100.0% | 100.0% |
| Total | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Bank account | 100.0% | 99.3% |
Documents where they differ
- Tesař s.r.o. · 20250237 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 10/13
- Holub s.r.o. · FV2500655 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Slavěna Pospíšilová · 26200409 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Agrotrade s.r.o. · DB-2025-00296 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Stavebniny Sýkora s.r.o. · FV-2025-00537 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Východočeská obchodní s.r.o. · FV2501176 · Photo · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
Invoice extraction · German (150 documents)
GPT-6 Luna is more accurate by 2 pp. GPT-6 Luna is 13.1× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 97.2% | 99.2% |
| Error-free documents | 96% | 91% |
| Per 1,000 documents | $7.15 | $0.54 |
| To fix / 1,000 | 40 | 87 |
| Speed (median) | 4.7 sec | 3.7 sec |
| Valid answers | 97.3% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Supplier | 97.3% | 100.0% |
| Supplier ID (EIN, IČO…) | 97.3% | 98.7% |
| VAT ID | 96.7% | 99.3% |
| Invoice number | 97.3% | 99.3% |
| Payment reference | 97.3% | 100.0% |
| Invoice date | 97.3% | 98.7% |
| Tax point | 97.3% | 100.0% |
| Due date | 97.3% | 100.0% |
| Net by tax rate | 96.7% | 100.0% |
| Tax by rate | 97.3% | 100.0% |
| Total | 97.3% | 100.0% |
| Currency | 97.3% | 100.0% |
| Bank account | 97.3% | 94.0% |
Documents where they differ
- Gebäudereinigung Klara Weiß · 2025-1126 · PDF text · Gemini 3.8 Flash 0/13, GPT-6 Luna 13/13
- Malerbetrieb Otfried Haas · RE-2025-00168 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Braun IT-Service GmbH · RE2500825 · Photo · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Wesermetall GmbH · RE2501375 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
Invoice extraction · Spanish (100 documents)
GPT-6 Luna is more accurate by 3.2 pp. GPT-6 Luna is 14.2× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 95.9% | 99.1% |
| Error-free documents | 95% | 88% |
| Per 1,000 documents | $7.86 | $0.55 |
| To fix / 1,000 | 50 | 120 |
| Speed (median) | 5.3 sec | 3.9 sec |
| Valid answers | 96.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Supplier | 95.0% | 99.0% |
| Supplier ID (EIN, IČO…) | 96.0% | 100.0% |
| VAT ID | 96.0% | 100.0% |
| Invoice number | 96.0% | 100.0% |
| Payment reference | 96.0% | 98.0% |
| Invoice date | 96.0% | 100.0% |
| Tax point | 96.0% | 100.0% |
| Due date | 96.0% | 100.0% |
| Net by tax rate | 96.0% | 100.0% |
| Tax by rate | 96.0% | 100.0% |
| Total | 96.0% | 100.0% |
| Currency | 96.0% | 100.0% |
| Bank account | 96.0% | 91.0% |
Documents where they differ
- Climatización Levante S.L. · F2025-1305 · Scan · Gemini 3.8 Flash 0/13, GPT-6 Luna 13/13
- Centro de Formación del Sur S.L.U. · F2025-1424 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Academia Checa S.L. · F2025-0208 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Gráficas Levante S.L.U. · F2025-0221 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
Invoice extraction · French (100 documents)
Gemini 3.8 Flash is more accurate by 1 pp. GPT-6 Luna is 10.9× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.9% | 98.9% |
| Error-free documents | 99% | 86% |
| Per 1,000 documents | $6.18 | $0.57 |
| To fix / 1,000 | 10 | 143 |
| Speed (median) | 5.2 sec | 3.8 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Supplier | 100.0% | 99.0% |
| Supplier ID (EIN, IČO…) | 100.0% | 99.0% |
| VAT ID | 100.0% | 98.0% |
| Invoice number | 100.0% | 99.0% |
| Payment reference | 99.0% | 100.0% |
| Invoice date | 100.0% | 100.0% |
| Tax point | 100.0% | 100.0% |
| Due date | 100.0% | 100.0% |
| Net by tax rate | 100.0% | 100.0% |
| Tax by rate | 100.0% | 100.0% |
| Total | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Bank account | 100.0% | 90.8% |
Documents where they differ
- Nettoyage Mercier SASU · AC-10176 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Carrosserie Gauthier SASU · F-11079 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Établissements Denis SASU · FAC-2025-00031 · Scan · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Carrosserie Alain Meunier EI · FAC-2026-00869 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Catherine Dubois EI · FAC-2025-00139 · PDF text · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- SAS Thomas & Fils · FAC-2025-01058 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
- Gilbert Marie – Multiservices (EI) · FAC-2025-00924 · Clean image · Gemini 3.8 Flash 13/13, GPT-6 Luna 12/13
Email triage · English (US) (150 documents)
Gemini 3.8 Flash is more accurate by 1.3 pp. GPT-6 Luna is 17.7× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 98.7% | 97.5% |
| Error-free documents | 89% | 77% |
| Per 1,000 documents | $2.39 | $0.13 |
| To fix / 1,000 | 113 | 227 |
| Speed (median) | 2.2 sec | 2.5 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Category | 100.0% | 100.0% |
| Priority | 100.0% | 98.0% |
| Sentiment | 88.7% | 79.3% |
| Customer name | 100.0% | 100.0% |
| Order number | 100.0% | 100.0% |
| Invoice number | 100.0% | 100.0% |
| Amount | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Deadline | 100.0% | 100.0% |
Documents where they differ
- Omar Cabrera · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Cody Baldwin · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Daniel Stewart · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Allison Floyd · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Scott Davis · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Sandra Hogan · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
Email triage · Czech (150 documents)
Gemini 3.8 Flash is more accurate by 1.3 pp. GPT-6 Luna is 17.2× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.2% | 97.9% |
| Error-free documents | 93% | 81% |
| Per 1,000 documents | $3.06 | $0.18 |
| To fix / 1,000 | 73 | 187 |
| Speed (median) | 2.6 sec | 2.6 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Category | 100.0% | 100.0% |
| Priority | 99.3% | 98.7% |
| Sentiment | 93.3% | 82.7% |
| Customer name | 100.0% | 100.0% |
| Order number | 100.0% | 100.0% |
| Invoice number | 100.0% | 100.0% |
| Amount | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Deadline | 100.0% | 100.0% |
Documents where they differ
- Simona Kolářová · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Václav Vítek · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Šimon Beran · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
Email triage · German (150 documents)
Gemini 3.8 Flash is more accurate by 1.4 pp. GPT-6 Luna is 18× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 98.4% | 97.0% |
| Error-free documents | 85% | 73% |
| Per 1,000 documents | $2.74 | $0.15 |
| To fix / 1,000 | 147 | 273 |
| Speed (median) | 2.9 sec | 2.8 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Category | 100.0% | 100.0% |
| Priority | 100.0% | 99.3% |
| Sentiment | 85.3% | 73.3% |
| Customer name | 100.0% | 100.0% |
| Order number | 100.0% | 100.0% |
| Invoice number | 100.0% | 100.0% |
| Amount | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Deadline | 100.0% | 100.0% |
Documents where they differ
- Alexej Sauer · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Sylwia Gerlach · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Irmela Warmer · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Linda Bähr · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Nadja Wiek · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Ibrahim Walter · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
- Margret Löchel · Email · Gemini 3.8 Flash 9/9, GPT-6 Luna 8/9
Personal data detection · English (US) (150 documents)
Gemini 3.8 Flash is more accurate by 0.1 pp. GPT-6 Luna is 17.6× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.7% | 99.6% |
| Error-free documents | 98% | 97% |
| Per 1,000 documents | $1.86 | $0.11 |
| To fix / 1,000 | 20 | 27 |
| Speed (median) | 1.9 sec | 1.8 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Names | 100.0% | 98.7% |
| Email addresses | 100.0% | 100.0% |
| Phone numbers | 100.0% | 100.0% |
| Addresses | 98.0% | 98.7% |
| Dates of birth | 100.0% | 100.0% |
| National IDs | 100.0% | 100.0% |
| Bank accounts | 100.0% | 100.0% |
Documents where they differ
- Justin Martinez · Text · Gemini 3.8 Flash 7/7, GPT-6 Luna 6/7
- Margaret Mills · Text · Gemini 3.8 Flash 7/7, GPT-6 Luna 6/7
Personal data detection · Czech (150 documents)
GPT-6 Luna is more accurate by 0.1 pp. GPT-6 Luna is 18.3× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.5% | 99.6% |
| Error-free documents | 97% | 97% |
| Per 1,000 documents | $2.19 | $0.12 |
| To fix / 1,000 | 33 | 27 |
| Speed (median) | 2.0 sec | 1.8 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Names | 98.7% | 100.0% |
| Email addresses | 100.0% | 100.0% |
| Phone numbers | 100.0% | 100.0% |
| Addresses | 98.0% | 97.3% |
| Dates of birth | 100.0% | 100.0% |
| National IDs | 100.0% | 100.0% |
| Bank accounts | 100.0% | 100.0% |
Documents where they differ
- Jonáš Toman · Text · Gemini 3.8 Flash 7/7, GPT-6 Luna 6/7
Personal data detection · German (150 documents)
Gemini 3.8 Flash is more accurate by 0.1 pp. GPT-6 Luna is 18.1× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.7% | 99.6% |
| Error-free documents | 98% | 97% |
| Per 1,000 documents | $2.12 | $0.12 |
| To fix / 1,000 | 20 | 27 |
| Speed (median) | 2.3 sec | 2.2 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Names | 100.0% | 100.0% |
| Email addresses | 100.0% | 100.0% |
| Phone numbers | 100.0% | 100.0% |
| Addresses | 98.0% | 97.3% |
| Dates of birth | 100.0% | 100.0% |
| National IDs | 100.0% | 100.0% |
| Bank accounts | 100.0% | 100.0% |
Documents where they differ
- Gunnar Zorbach · Text · Gemini 3.8 Flash 7/7, GPT-6 Luna 6/7
Contract clauses · English (US) (150 documents)
Gemini 3.8 Flash is more accurate by 0.1 pp. GPT-6 Luna is 12.3× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.9% | 99.8% |
| Error-free documents | 99% | 99% |
| Per 1,000 documents | $2.47 | $0.20 |
| To fix / 1,000 | 7 | 13 |
| Speed (median) | 2.0 sec | 2.7 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Provider | 99.3% | 98.7% |
| Provider ID | 100.0% | 100.0% |
| Client | 99.3% | 98.7% |
| Client ID | 100.0% | 100.0% |
| Contract type | 100.0% | 100.0% |
| Price (net) | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Billing period | 100.0% | 100.0% |
| Effective date | 100.0% | 100.0% |
| End date | 100.0% | 100.0% |
| Notice period (months) | 100.0% | 100.0% |
| Late-payment penalty (% per day) | 100.0% | 100.0% |
| Governing law | 100.0% | 100.0% |
Documents where they differ
- Pittsburgh Construction Co. · Evans Electric LLC · Contract text · Gemini 3.8 Flash 13/13, GPT-6 Luna 11/13
- Kim Electric Co. · Ramos & Sons, Co. · Contract text · Gemini 3.8 Flash 11/13, GPT-6 Luna 13/13
Contract clauses · Czech (150 documents)
Both are equally accurate. GPT-6 Luna is 11.4× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.0% | 100.0% |
| Error-free documents | 100% | 100% |
| Per 1,000 documents | $3.09 | $0.27 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 2.2 sec | 2.6 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Provider | 100.0% | 100.0% |
| Provider ID | 100.0% | 100.0% |
| Client | 100.0% | 100.0% |
| Client ID | 100.0% | 100.0% |
| Contract type | 100.0% | 100.0% |
| Price (net) | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Billing period | 100.0% | 100.0% |
| Effective date | 100.0% | 100.0% |
| End date | 100.0% | 100.0% |
| Notice period (months) | 100.0% | 100.0% |
| Late-payment penalty (% per day) | 100.0% | 100.0% |
| Governing law | 100.0% | 100.0% |
Documents where they differ
Both models got the same number of fields right on every example document.
Contract clauses · German (150 documents)
GPT-6 Luna is more accurate by 0.1 pp. GPT-6 Luna is 12.8× cheaper.
| Metric | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.9% | 100.0% |
| Error-free documents | 99% | 100% |
| Per 1,000 documents | $3.42 | $0.27 |
| To fix / 1,000 | 7 | 0 |
| Speed (median) | 2.6 sec | 3.0 sec |
| Valid answers | 100.0% | 100.0% |
Accuracy per field
| Field | Gemini 3.8 Flash | GPT-6 Luna |
|---|---|---|
| Provider | 100.0% | 100.0% |
| Provider ID | 99.3% | 100.0% |
| Client | 100.0% | 100.0% |
| Client ID | 99.3% | 100.0% |
| Contract type | 100.0% | 100.0% |
| Price (net) | 100.0% | 100.0% |
| Currency | 100.0% | 100.0% |
| Billing period | 100.0% | 100.0% |
| Effective date | 100.0% | 100.0% |
| End date | 100.0% | 100.0% |
| Notice period (months) | 100.0% | 100.0% |
| Late-payment penalty (% per day) | 100.0% | 100.0% |
| Governing law | 100.0% | 100.0% |
Documents where they differ
Both models got the same number of fields right on every example document.