DeepSeek V4 Flash vs Qwen3 235B A22B Thinking 2507

DeepSeek V4 Flash and Qwen3 235B A22B Thinking 2507 match in llmrun Score, with values of 109.3 versus 108.1; neither fits 24 GB.

Model A · DeepSeek

DeepSeek V4 Flash

109.3llmrun Score

290.9B · 14.1B active · 1M context · MIT · Apr 2026

Chat

Model B · Alibaba

Qwen3 235B A22B Thinking 2507

108.1llmrun Score

235.1B · 22.2B active · 256K context · Apache 2.0 · Jul 2025

ChatFunctions

Which should you choose?

Choose DeepSeek V4 Flash if you want…

Ahead on the benchmarks
6 of 6 shared boards
Longer context window
1M vs 256K tokens
Newer release
Apr 2026 vs Jul 2025

Choose Qwen3 235B A22B Thinking 2507 if you want…

Lighter on memory
141.6 vs 174.9 GB at Q4_K_M
Also does function-calling
DeepSeek V4 Flash doesn't list it

On your hardware

Each model at its default quant (Q4_K_M), a smaller one when it must. Estimates, ±20%.

MemoryDeepSeek V4 FlashQwen3 235B A22B Thinking 2507

8 GB

RTX 4060 · 8 GB tier

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

12 GB

RTX 3060 12GB · 12 GB tier

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

16 GB

RTX 5060 Ti 16GB · 16 GB tier

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

24 GBreference

RTX 4090 · 24 GB tier

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

32 GB

RTX 5090 · 32 GB tier

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

48 GB

RTX A6000 · 48 GB tier

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

Mac 32 GB

Mac Mini M4 · Macs

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

Mac 64 GB

Mac Studio M4 Max · Macs

DeepSeek V4 Flash

Doesn't fit

smallest quant needs 80.5 GB

Qwen3 235B A22B Thinking 2507

Doesn't fit

smallest quant needs 72.4 GB

Mac 128 GB

Mac Studio M4 Max · Macs

DeepSeek V4 Flash

IQ3_S · ~41 tok/s

124.1 GB · up to 32K

Qwen3 235B A22B Thinking 2507

Q3_K_L · ~21 tok/s

122.4 GB · up to 32K

Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).

llmrun Score

with its 90% range · bars share one scale

DeepSeek V4 Flash Qwen3 235B A22B Thinking 2507

Overall

DeepSeek V4 Flash

109.3104–114

Qwen3 235B A22B Thinking 2507

108.1105–112

Coding

DeepSeek V4 Flash

106.699–115

Qwen3 235B A22B Thinking 2507

102.693–113

Science

DeepSeek V4 Flash

111.492–130

Qwen3 235B A22B Thinking 2507

107.799–116

Reasoning

DeepSeek V4 Flash

111.8105–119

Qwen3 235B A22B Thinking 2507

105.998–113

DeepSeek V4 Flash · #26 of 92 open models · #12 of 71 that fit ≤192 GB

Qwen3 235B A22B Thinking 2507 · #31 of 92 open models · #15 of 71 that fit ≤192 GB

Benchmarks

6 shared boards: DeepSeek V4 Flash leads 6, Qwen3 235B A22B Thinking 2507 leads 0 · Benchmarks imported Oct 2026

BoardDeepSeek V4 FlashQwen3 235B A22B Thinking 2507

LMArena Text

DeepSeek V4 Flash

1438.6#21/186 open

ArenaLMArena · Oct 2026

Qwen3 235B A22B Thinking 2507

1400.0#48/186 open

ArenaLMArena · Oct 2026

SciCode

DeepSeek V4 Flash

44.9%#21/65 open

Lab-reportedEpoch AI · Apr 2026

Qwen3 235B A22B Thinking 2507

42.4%#26/65 open

Lab-reportedEpoch AI · Jul 2025

WeirdML

DeepSeek V4 Flash

45.6%#14/46 open

IndependentEpoch AI · Apr 2026

Qwen3 235B A22B Thinking 2507

41.0%#20/46 open

IndependentEpoch AI · Jul 2025

CritPt

DeepSeek V4 Flash

7.1%#14/65 open

Lab-reportedEpoch AI · Apr 2026

Qwen3 235B A22B Thinking 2507

0.0%#62/65 open

Lab-reportedEpoch AI · Jul 2025

DTBench

DeepSeek V4 Flash

86.4%#14/68 open

Lab-reportedEpoch AI · Apr 2026

Qwen3 235B A22B Thinking 2507

80.3%#22/68 open

Lab-reportedEpoch AI · Jul 2025

LMCA

DeepSeek V4 Flash

35.9%#14/54 open

Lab-reportedEpoch AI · Apr 2026

Qwen3 235B A22B Thinking 2507

29.3%#23/54 open

Lab-reportedEpoch AI · Jul 2025

14 more boards (other shared and single-model boards)

Only DeepSeek V4 Flash (6)

Only Qwen3 235B A22B Thinking 2507 (8)

VRAM by quantization

Memory to load the model and 2K context; below it, the total at 32K and 128K context. Lighter side in bold.

QuantDeepSeek V4 FlashQwen3 235B A22B Thinking 2507

Q2_K

DeepSeek V4 Flash

124 GB107.0 GB file

32K 124 GB · 128K 125 GB

Qwen3 235B A22B Thinking 2507

100 GB85.7 GB file

32K 103 GB · 128K 113 GB

Q3_K_S

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

103 GB101.4 GB file

32K 106 GB · 128K 116 GB

Q3_K_M

DeepSeek V4 Flash

142 GB129.3 GB file

32K 143 GB · 128K 144 GB

Qwen3 235B A22B Thinking 2507

115 GB112.4 GB file

32K 118 GB · 128K 128 GB

Q4_0

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

118 GB133.1 GB file

32K 121 GB · 128K 130 GB

Q4_K_M

DeepSeek V4 Flash

175 GB174.9 GB file

32K 175 GB · 128K 176 GB

Qwen3 235B A22B Thinking 2507

142 GB142.2 GB file

32K 145 GB · 128K 154 GB

Q5_K_M

DeepSeek V4 Flash

208 GB~207.3 GB file

32K 208 GB · 128K 209 GB

Qwen3 235B A22B Thinking 2507

168 GB166.8 GB file

32K 171 GB · 128K 180 GB

Q6_K

DeepSeek V4 Flash

240 GB~240.0 GB file

32K 241 GB · 128K 242 GB

Qwen3 235B A22B Thinking 2507

194 GB193.0 GB file

32K 197 GB · 128K 207 GB

Q8_0

DeepSeek V4 Flash

291 GB302.3 GB file

32K 292 GB · 128K 293 GB

Qwen3 235B A22B Thinking 2507

236 GB249.9 GB file

32K 239 GB · 128K 248 GB

16 more quants

IQ2_XXS

DeepSeek V4 Flash

80.3 GB90.9 GB file

32K 80.7 GB · 128K 81.8 GB

Qwen3 235B A22B Thinking 2507

not listed

IQ2_XS

DeepSeek V4 Flash

87.6 GB87.0 GB file

32K 88.0 GB · 128K 89.0 GB

Qwen3 235B A22B Thinking 2507

71.0 GB65.6 GB file

32K 74.0 GB · 128K 83.5 GB

IQ2_S

DeepSeek V4 Flash

91.2 GB88.0 GB file

32K 91.6 GB · 128K 92.7 GB

Qwen3 235B A22B Thinking 2507

74.0 GB66.1 GB file

32K 76.9 GB · 128K 86.4 GB

IQ2_M

DeepSeek V4 Flash

98.5 GB90.9 GB file

32K 98.9 GB · 128K 99.9 GB

Qwen3 235B A22B Thinking 2507

79.8 GB78.6 GB file

32K 82.8 GB · 128K 92.3 GB

IQ3_XXS

DeepSeek V4 Flash

113 GB103.0 GB file

32K 113 GB · 128K 114 GB

Qwen3 235B A22B Thinking 2507

91.6 GB93.1 GB file

32K 94.6 GB · 128K 104 GB

Q2_K_S

DeepSeek V4 Flash

117 GB98.6 GB file

32K 117 GB · 128K 118 GB

Qwen3 235B A22B Thinking 2507

not listed

IQ3_XS

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

97.5 GB96.9 GB file

32K 100 GB · 128K 110 GB

IQ3_S

DeepSeek V4 Flash

124 GB117.3 GB file

32K 124 GB · 128K 125 GB

Qwen3 235B A22B Thinking 2507

not listed

IQ3_M

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

106 GB107.3 GB file

32K 109 GB · 128K 119 GB

Q3_K_L

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

121 GB111.2 GB file

32K 124 GB · 128K 133 GB

IQ4_XS

DeepSeek V4 Flash

157 GB137.9 GB file

32K 157 GB · 128K 158 GB

Qwen3 235B A22B Thinking 2507

127 GB125.5 GB file

32K 130 GB · 128K 139 GB

IQ4_NL

DeepSeek V4 Flash

164 GB137.9 GB file

32K 164 GB · 128K 165 GB

Qwen3 235B A22B Thinking 2507

133 GB133.1 GB file

32K 136 GB · 128K 145 GB

Q4_1

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

133 GB147.2 GB file

32K 136 GB · 128K 145 GB

Q4_K_S

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

133 GB133.7 GB file

32K 136 GB · 128K 145 GB

Q5_K_S

DeepSeek V4 Flash

not listed

Qwen3 235B A22B Thinking 2507

162 GB161.9 GB file

32K 165 GB · 128K 175 GB

BF16

DeepSeek V4 Flash

582 GB~581.9 GB file

32K 583 GB · 128K 584 GB

Qwen3 235B A22B Thinking 2507

471 GB470.3 GB file

32K 474 GB · 128K 483 GB

How to run each

DeepSeek V4 Flash

Ollama · Q4_K_M

$ollama run hf.co/teamblobfish/DeepSeek-V4-Flash-GGUF:Q4_K_M

All ways to run DeepSeek V4 Flash →

Qwen3 235B A22B Thinking 2507

Ollama · Q4_K_M

$ollama run hf.co/unsloth/Qwen3-235B-A22B-Thinking-2507-GGUF:Q4_K_M

All ways to run Qwen3 235B A22B Thinking 2507 →

Specifications

Specifications of DeepSeek V4 Flash and Qwen3 235B A22B Thinking 2507
Spec	DeepSeek V4 Flash	Qwen3 235B A22B Thinking 2507
Publisher	DeepSeek	Alibaba
Parameters	290.9B	235.1B
Active parameters	14B per token	22B per token
Architecture	DeepseekV4ForCausalLM	Qwen3MoeForCausalLM
Context	1M tokens	256K tokens
Modalities	Chat	Chat, Function-calling
License	MIT	Apache 2.0
Released	Apr 2026	Jul 2025

Frequently Asked Questions

Which is better, DeepSeek V4 Flash or Qwen3 235B A22B Thinking 2507?

DeepSeek V4 Flash and Qwen3 235B A22B Thinking 2507 match in llmrun Score, with values of 109.3 versus 108.1; neither fits 24 GB. Choose DeepSeek V4 Flash for ahead on the benchmarks and longer context window; choose Qwen3 235B A22B Thinking 2507 for lighter on memory and also does function-calling.

Which needs less VRAM, DeepSeek V4 Flash or Qwen3 235B A22B Thinking 2507?

At Q4_K_M, DeepSeek V4 Flash requires 174.9 GB while Qwen3 235B A22B Thinking 2507 needs 141.6 GB, making Qwen3 235B A22B Thinking 2507 the easier model to run on your local setup.

Can I run DeepSeek V4 Flash or Qwen3 235B A22B Thinking 2507 on a 16 GB GPU?

DeepSeek V4 Flash does not fit (smallest quant needs 80.5 GB); Qwen3 235B A22B Thinking 2507 does not fit (smallest quant needs 72.4 GB).

Which has a longer context window, DeepSeek V4 Flash or Qwen3 235B A22B Thinking 2507?

DeepSeek V4 Flash supports 1,048,576 tokens and Qwen3 235B A22B Thinking 2507 supports 262,144 tokens.