AI BENCHY Compare
DeepSeek: DeepSeek V3.2 vs StepFun: Step 3.5 Flash
Benchmark zimetengenezwa kutoka seti za majaribio za AI BENCHY tarehe: 2026-03-03
| Kipimo | DeepSeek: DeepSeek V3.2 none Toleo: 2025-12-01 | StepFun: Step 3.5 Flash medium Toleo: 2026-02-01 Inapatikana bure |
|---|---|---|
| Nafasi | #30 | #11 |
| Wastani wa alama | 4.70 | 7.00 |
| Uthabiti | 8.19 | 8.32 |
| Gharama kwa matokeo | 0.132 | 0.000 |
| Jumla ya gharama | $0.007 | $0.000 |
| Majaribio sahihi | 5/14 | 9/14 |
| Kiwango cha kupita kwa kila jaribio | 47.6% | 73.8% |
| Majaribio yasiyo thabiti | 3 | 3 |
| Tokeni za matokeo | 4,869 | 60,502 |
| Tokeni za hoja | 0 | 117,044 |
Mgawanyo wa kategoria
| Mbinu za kupinga AI | Alama | Uthabiti | Kiwango cha kupita kwa kila jaribio | Majaribio yasiyo thabiti | Majaribio sahihi | Tokeni za matokeo | Tokeni za hoja |
|---|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V3.2 | 1.00 | 9.68 | 0.0% | 0 | 1,411 | 0 | |
| StepFun: Step 3.5 Flash | 10.00 | 10.00 | 100.0% | 0 | 13,924 | 17,208 |
| Uchanganuzi na uchimbaji wa data | Alama | Uthabiti | Kiwango cha kupita kwa kila jaribio | Majaribio yasiyo thabiti | Majaribio sahihi | Tokeni za matokeo | Tokeni za hoja |
|---|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V3.2 | 5.38 | 5.81 | 66.7% | 1 | 1,710 | 0 | |
| StepFun: Step 3.5 Flash | 10.00 | 10.00 | 100.0% | 0 | 535 | 11,548 |
| Mahususi kwa domeni | Alama | Uthabiti | Kiwango cha kupita kwa kila jaribio | Majaribio yasiyo thabiti | Majaribio sahihi | Tokeni za matokeo | Tokeni za hoja |
|---|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V3.2 | 1.00 | 7.21 | 22.2% | 1 | 24 | 0 | |
| StepFun: Step 3.5 Flash | 4.00 | 7.21 | 44.4% | 1 | 40,942 | 74,237 |
| Ufuataji wa maagizo | Alama | Uthabiti | Kiwango cha kupita kwa kila jaribio | Majaribio yasiyo thabiti | Majaribio sahihi | Tokeni za matokeo | Tokeni za hoja |
|---|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V3.2 | 8.00 | 9.99 | 50.0% | 0 | 66 | 0 | |
| StepFun: Step 3.5 Flash | 10.00 | 10.00 | 100.0% | 0 | 2,121 | 3,274 |
| Puzzle Solving | Alama | Uthabiti | Kiwango cha kupita kwa kila jaribio | Majaribio yasiyo thabiti | Majaribio sahihi | Tokeni za matokeo | Tokeni za hoja |
|---|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V3.2 | 7.67 | 7.49 | 88.9% | 1 | 1,136 | 0 | |
| StepFun: Step 3.5 Flash | 2.00 | 4.96 | 33.3% | 2 | 2,705 | 6,975 |
| Mwito wa zana | Alama | Uthabiti | Kiwango cha kupita kwa kila jaribio | Majaribio yasiyo thabiti | Majaribio sahihi | Tokeni za matokeo | Tokeni za hoja |
|---|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V3.2 | 10.00 | 10.00 | 100.0% | 0 | 522 | 0 | |
| StepFun: Step 3.5 Flash | 10.00 | 10.00 | 100.0% | 0 | 275 | 3,802 |
Ulinganisho wa haraka
Badilisha jozi ya ulinganisho
Gemini 3 Flash PreviewnonevsStep 3.5 FlashmediumInapatikana bureDeepSeek V3.2nonevsGLM 4.7 FlashmediumClaude Sonnet 4.6nonevsStep 3.5 FlashmediumInapatikana bureQwen3.5 Plus 2026-02-15nonevsStep 3.5 FlashmediumInapatikana bureStep 3.5 FlashmediumInapatikana burevsGLM 5noneDeepSeek V3.2nonevsMiniMax M2.5mediumDeepSeek V3.2nonevsgpt-oss-120bmediumInapatikana bureDeepSeek V3.2nonevsQwen3.5-FlashmediumDeepSeek V3.2nonevsGPT-5 NanomediumDeepSeek V3.2nonevsQwen3 Coder NextmediumDeepSeek V3.2nonevsQwen3.5-35B-A3BmediumDeepSeek V3.2nonevsMiMo-V2-Flashmedium