DeepSeek V4 Flash 0731 deepseek-v4-flash-0731

deepseek-ai · deepseek reasoning efficient

unranked — no quality signals yet

params
158B
arch
moe
context
1024k
license
mit open weights
released
Jul 2026
reasoning
yes
downloads/30d
3.2M

# architecture

attn shared · reasoning GQA 64q/1kv router top-6 of 256 +1 shared expert ×256 expert out ×43 layers ctx 1,048,576 158B pool
layers
43
d_model
4096
heads
64q / 1kv
head dim
512
experts
top-6 of 256 + 1 shared
vocab
129280
family
deepseek_v4

signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.

# trend

OGM score · last 92 days
OGM score over 92 days
overall rank (up is better)
Overall rank over 92 days

methodology changed during this history window; score movement across that boundary is not model movement. See methodology v6.

# benchmarks

benchmarkscoresourcedate
Agents-last-exam 0.3 / 1 LLM Stats —
Automationbench 0.3 / 1 LLM Stats —

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime✓
Featherless — ——— via hfrouter — —
Nebius — ——— nebius — — ✓
Scaleway — ——— via hfrouter 145 —
Relacefp4 1024k $0.07$0.14$0.01 via openrouter — 100.0%
DeepInfrafp8 1024k $0.06$0.18— deepinfra — 100.0% ✓
Sail 1024k $0.09$0.18— via litellm — —
Wafernvfp4 1024k $0.06$0.30$0.04 via openrouter — 100.0%
StreamLakefp8 1000k $0.09$0.26$0.00 via openrouter — 100.0%
Vultr 1024k $0.10$0.25— via modelsdev — —
Rekaunknown 256k $0.10$0.28$0.04 via openrouter — 99.9%
DigitalOceanunknown 1024k $0.12$0.24$0.02 via openrouter — 100.0%
BaseTenfp8 1024k $0.13$0.26$0.03 via openrouter — 100.0%
BaseTenfp32 1024k $0.13$0.26$0.03 via openrouter — 100.0%
Veniceunknown 1000k $0.13$0.26— venice — 99.7% ✓
Wafernvfp4 1024k $0.14$0.23$0.14 via openrouter — 100.0%
CoreWeavefp8 256k $0.13$0.28$0.07 via openrouter — 100.0%
Wandb 262k $0.13$0.28— via litellm — —
Cohereunknown 1024k $0.14$0.28$0.07 via openrouter — 100.0%
Parasailfp8 1024k $0.14$0.28$0.05 via openrouter — 100.0%
Together 1024k $0.14$0.28$0.03 via modelsdev — —
Sail Researchfp4 1024k $0.15$0.30$0.04 via openrouter — 100.0%
Sail Researchfp4 1024k $0.15$0.30$0.04 via openrouter — 100.0%
Inceptronfp4 1024k $0.13$0.49$0.03 via openrouter — 100.0%
Inferenceapis 1024k $0.19$0.38— via litellm — —
Qwen Ai Platform 1000k $0.20$0.40— via litellm — —
Qwencloud 1000k $0.20$0.40— via litellm — —
Alibabaunknown 1000k $0.18$0.53$0.02 via openrouter — 99.6%
Alibabaunknown 1000k $0.19$0.56$0.02 via openrouter — 99.9%
Mancer 2fp8 1024k $0.20$0.60— via openrouter — 100.0%
Fireworks 1000k $0.22$0.66$0.01 via requesty — —
SiliconFlowfp8 1024k $0.22$0.66$0.03 openrouter — 99.8% ✓
GMICloudfp8 1048k $0.29$0.86$0.01 via openrouter — 99.6%
AtlasCloudfp4 1024k $0.44$1.32— atlascloud — 100.0% ✓
Azure 1000k $0.44$1.32— via litellm — —
Baidufp8 1024k $0.44$1.32$0.01 via openrouter — 100.0%
Chutesfp8 1024k $0.44$1.32$0.04 via openrouter — 100.0%
Cloudflareunknown 1024k $0.44$1.32$0.01 via openrouter — 100.0%
Novitafp8 1024k $0.44$1.32— novita — 100.0% ✓
Phalaunknown 1024k $0.44$1.32— phala — 100.0% ✓

sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price

source aliases
aa
deepseek-v4-flash, deepseek-v4-flash-non-reasoning
hf
bartowski/DeepSeek-V4-Flash-0731-GGUF, unsloth/DeepSeek-V4-Flash-0731
openrouter
deepseek/deepseek-v4-flash-0731