ModelVerdict

Comparison

DeepSeek V4.1 Flash vs Gemma 4 31B

Same documents, same prompt: where each model is right, what it really costs and how fast it answers.

At a glance

Task · documentsEnglish (US)CzechGermanSpanishFrench
Invoice extraction99.4% / 97.6%99.9% / 96.1%99.5% / 94.9%98.3% / 95.4%99.4% / 93.6%
Email triage97.6% / 96.5%97.4% / 97.1%96.3% / 95.1%——
Personal data detection99.6% / 99.9%99.7% / 99.8%99.8% / 99.9%——
Contract clauses100.0% / 99.6%100.0% / 100.0%97.2% / 94.4%——

Correct fields: DeepSeek V4.1 Flash / Gemma 4 31B; the better one in bold. Click a cell to see the details.

Invoice extraction · English (US) (150 documents)

DeepSeek V4.1 Flash is more accurate by 1.7 pp. Gemma 4 31B is 3.1× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.4%97.6%
Error-free documents93%81%
Per 1,000 documents$0.52$0.17
To fix / 1,00067187
Speed (median)5.7 sec8.9 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Supplier100.0%99.3%
Supplier ID (EIN, IČO…)100.0%98.0%
Invoice number100.0%100.0%
Payment reference96.7%99.3%
Invoice date100.0%98.7%
Due date100.0%98.7%
Net by tax rate100.0%95.3%
Tax by rate100.0%98.7%
Total100.0%100.0%
Currency99.3%100.0%
Bank account97.3%86.0%

Documents where they differ

Invoice extraction · Czech (150 documents)

DeepSeek V4.1 Flash is more accurate by 3.8 pp. Gemma 4 31B is 2.3× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.9%96.1%
Error-free documents99%67%
Per 1,000 documents$0.50$0.21
To fix / 1,00013333
Speed (median)5.0 sec10.4 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Supplier99.3%96.7%
Supplier ID (EIN, IČO…)100.0%93.3%
VAT ID100.0%96.7%
Invoice number100.0%100.0%
Payment reference100.0%96.0%
Invoice date100.0%96.0%
Tax point100.0%95.3%
Due date100.0%94.0%
Net by tax rate100.0%98.0%
Tax by rate100.0%97.3%
Total100.0%99.3%
Currency100.0%100.0%
Bank account99.3%86.7%

Documents where they differ

Invoice extraction · German (150 documents)

DeepSeek V4.1 Flash is more accurate by 4.7 pp. Gemma 4 31B is 2.8× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.5%94.9%
Error-free documents94%64%
Per 1,000 documents$0.57$0.20
To fix / 1,00060360
Speed (median)6.4 sec7.7 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Supplier99.3%98.7%
Supplier ID (EIN, IČO…)98.7%99.3%
VAT ID99.3%94.0%
Invoice number100.0%99.3%
Payment reference100.0%98.7%
Invoice date100.0%87.3%
Tax point99.3%92.0%
Due date99.3%90.7%
Net by tax rate100.0%94.7%
Tax by rate100.0%97.3%
Total100.0%99.3%
Currency100.0%99.3%
Bank account98.0%82.7%

Documents where they differ

Invoice extraction · Spanish (79 documents)

DeepSeek V4.1 Flash is more accurate by 2.9 pp. Gemma 4 31B is 3.5× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields98.3%95.4%
Error-free documents77%54%
Per 1,000 documents$0.71$0.20
To fix / 1,000228460
Speed (median)5.8 sec8.7 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Supplier93.7%95.4%
Supplier ID (EIN, IČO…)100.0%93.1%
VAT ID100.0%98.9%
Invoice number100.0%100.0%
Payment reference97.5%95.4%
Invoice date100.0%98.9%
Tax point98.7%100.0%
Due date100.0%97.7%
Net by tax rate88.6%81.6%
Tax by rate100.0%100.0%
Total100.0%100.0%
Currency100.0%100.0%
Bank account98.7%79.3%

Documents where they differ

Invoice extraction · French (74 documents)

DeepSeek V4.1 Flash is more accurate by 5.8 pp. Gemma 4 31B is 2.9× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.4%93.6%
Error-free documents92%51%
Per 1,000 documents$0.65$0.22
To fix / 1,00081494
Speed (median)5.7 sec7.2 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Supplier100.0%96.6%
Supplier ID (EIN, IČO…)100.0%77.0%
VAT ID98.7%95.4%
Invoice number100.0%97.7%
Payment reference98.7%96.6%
Invoice date100.0%96.6%
Tax point100.0%94.3%
Due date100.0%96.6%
Net by tax rate100.0%93.1%
Tax by rate100.0%100.0%
Total100.0%100.0%
Currency100.0%98.9%
Bank account94.6%73.6%

Documents where they differ

Email triage · English (US) (150 documents)

DeepSeek V4.1 Flash is more accurate by 1.1 pp. Gemma 4 31B is 1.9× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields97.6%96.5%
Error-free documents79%72%
Per 1,000 documents$0.23$0.12
To fix / 1,000213280
Speed (median)3.8 sec2.2 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Category100.0%100.0%
Priority96.7%85.3%
Sentiment82.0%83.3%
Customer name100.0%100.0%
Order number100.0%100.0%
Invoice number100.0%100.0%
Amount100.0%100.0%
Currency100.0%100.0%
Deadline100.0%100.0%

Documents where they differ

  • Allison Floyd · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 7/9
  • Russell Bates · Email · DeepSeek V4.1 Flash 8/9, Gemma 4 31B 9/9
  • Sheila Evans · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Gene Everett · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Laurie Knight · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Jason Davis · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Daniel Stewart · Email · DeepSeek V4.1 Flash 8/9, Gemma 4 31B 7/9
  • Christine Copeland · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9

Email triage · Czech (150 documents)

DeepSeek V4.1 Flash is more accurate by 0.3 pp. Gemma 4 31B is 1.9× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields97.4%97.1%
Error-free documents77%75%
Per 1,000 documents$0.30$0.16
To fix / 1,000227253
Speed (median)4.2 sec2.5 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Category100.0%100.0%
Priority98.7%85.3%
Sentiment78.0%88.7%
Customer name100.0%100.0%
Order number100.0%100.0%
Invoice number100.0%100.0%
Amount100.0%100.0%
Currency100.0%100.0%
Deadline100.0%100.0%

Documents where they differ

Email triage · German (145 documents)

DeepSeek V4.1 Flash is more accurate by 1.2 pp. Gemma 4 31B is 2× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields96.3%95.1%
Error-free documents68%62%
Per 1,000 documents$0.30$0.15
To fix / 1,000324380
Speed (median)4.2 sec3.6 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Category100.0%100.0%
Priority97.2%78.0%
Sentiment69.7%78.0%
Customer name100.0%100.0%
Order number100.0%100.0%
Invoice number100.0%100.0%
Amount100.0%100.0%
Currency100.0%100.0%
Deadline100.0%100.0%

Documents where they differ

  • Irmela Warmer · Email · DeepSeek V4.1 Flash 8/9, Gemma 4 31B 9/9
  • Cristina Krause · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Cynthia Hahn · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Linda Bähr · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Sibille Lehmann · Email · DeepSeek V4.1 Flash 8/9, Gemma 4 31B 9/9
  • Nadja Wiek · Email · DeepSeek V4.1 Flash 8/9, Gemma 4 31B 9/9
  • Marta Kambs · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9
  • Samira Hermann · Email · DeepSeek V4.1 Flash 9/9, Gemma 4 31B 8/9

Personal data detection · English (US) (150 documents)

Gemma 4 31B is more accurate by 0.3 pp. Gemma 4 31B is 3.4× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.6%99.9%
Error-free documents97%99%
Per 1,000 documents$0.29$0.08
To fix / 1,000277
Speed (median)3.8 sec2.4 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Names98.7%100.0%
Email addresses100.0%100.0%
Phone numbers100.0%100.0%
Addresses98.7%99.3%
Dates of birth100.0%100.0%
National IDs100.0%100.0%
Bank accounts100.0%100.0%

Documents where they differ

Personal data detection · Czech (150 documents)

Gemma 4 31B is more accurate by 0.1 pp. Gemma 4 31B is 2.6× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.7%99.8%
Error-free documents98%99%
Per 1,000 documents$0.31$0.12
To fix / 1,0002013
Speed (median)4.2 sec2.9 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Names100.0%100.0%
Email addresses100.0%100.0%
Phone numbers100.0%100.0%
Addresses98.0%98.7%
Dates of birth100.0%100.0%
National IDs100.0%100.0%
Bank accounts100.0%100.0%

Documents where they differ

Both models got the same number of fields right on every example document.

Personal data detection · German (148 documents)

Gemma 4 31B is more accurate by 0.1 pp. Gemma 4 31B is 3× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields99.8%99.9%
Error-free documents99%99%
Per 1,000 documents$0.31$0.10
To fix / 1,000137
Speed (median)3.7 sec1.9 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Names100.0%100.0%
Email addresses100.0%100.0%
Phone numbers100.0%100.0%
Addresses98.7%99.3%
Dates of birth100.0%100.0%
National IDs100.0%100.0%
Bank accounts100.0%100.0%

Documents where they differ

Both models got the same number of fields right on every example document.

Contract clauses · English (US) (150 documents)

DeepSeek V4.1 Flash is more accurate by 0.4 pp. Gemma 4 31B is 1.8× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields100.0%99.6%
Error-free documents100%97%
Per 1,000 documents$0.32$0.18
To fix / 1,000027
Speed (median)3.9 sec2.5 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Provider100.0%98.0%
Provider ID100.0%100.0%
Client100.0%98.0%
Client ID100.0%100.0%
Contract type100.0%100.0%
Price (net)100.0%100.0%
Currency100.0%100.0%
Billing period100.0%100.0%
Effective date100.0%99.3%
End date100.0%100.0%
Notice period (months)100.0%100.0%
Late-payment penalty (% per day)100.0%100.0%
Governing law100.0%100.0%

Documents where they differ

Both models got the same number of fields right on every example document.

Contract clauses · Czech (150 documents)

Both are equally accurate. Gemma 4 31B is 1.7× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields100.0%100.0%
Error-free documents100%99%
Per 1,000 documents$0.42$0.25
To fix / 1,00007
Speed (median)4.3 sec6.0 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Provider100.0%100.0%
Provider ID100.0%100.0%
Client100.0%100.0%
Client ID100.0%100.0%
Contract type100.0%100.0%
Price (net)100.0%100.0%
Currency100.0%100.0%
Billing period100.0%100.0%
Effective date100.0%99.3%
End date100.0%100.0%
Notice period (months)100.0%100.0%
Late-payment penalty (% per day)100.0%100.0%
Governing law100.0%100.0%

Documents where they differ

Contract clauses · German (145 documents)

DeepSeek V4.1 Flash is more accurate by 2.8 pp. Gemma 4 31B is 2.3× cheaper.

MetricDeepSeek V4.1 FlashGemma 4 31B
Correct fields97.2%94.4%
Error-free documents77%27%
Per 1,000 documents$0.51$0.22
To fix / 1,000228727
Speed (median)4.6 sec2.6 sec
Valid answers100.0%100.0%

Accuracy per field

FieldDeepSeek V4.1 FlashGemma 4 31B
Provider100.0%100.0%
Provider ID81.4%100.0%
Client100.0%100.0%
Client ID82.8%100.0%
Contract type100.0%100.0%
Price (net)100.0%100.0%
Currency100.0%100.0%
Billing period100.0%100.0%
Effective date100.0%100.0%
End date100.0%100.0%
Notice period (months)100.0%100.0%
Late-payment penalty (% per day)100.0%27.3%
Governing law100.0%100.0%

Documents where they differ