Coding · Benchmark profile

DeepSeek DSBench FullStack

DeepSeek's internal full-stack coding-agent benchmark.

Reference2026Active8 models

Data verified 1 Aug 2026 · Methodology 2.2.0

Benchmark score on DeepSeek DSBench FullStack

Leaderboard · 8 models

Claude Fable 5 leads at 77.2%, followed by Kimi K3 (73.7%) and Claude Opus 4.8 (71.6%).

Display onlyCoding2026

#1Closed

Claude Fable 5

Anthropic

claude-fable-5

77.2%

Overall #2 · Context 1M tokens

#2Closed

Kimi K3

Moonshot AI

kimi-k3

73.7%

Overall #4 · Context 1M tokens

#3Closed

Claude Opus 4.8

Anthropic

claude-opus-4-8

71.6%

Overall #8 · Context 1M tokens

ViewModelsColours

AnthropicMoonshot AIDeepSeekZ.AI

Lumina

DeepSeek DSBench FullStack leaderboard

Accessible data for DeepSeek DSBench FullStack leaderboard
Model	Rank	Model ID	Provider	Score	Unit
Claude Fable 5	#1	claude-fable-5	Anthropic	77.2	percent
Kimi K3	#2	kimi-k3	Moonshot AI	73.7	percent
Claude Opus 4.8	#3	claude-opus-4-8	Anthropic	71.6	percent
DeepSeek V4 Pro 0813	#4	deepseek-v4-pro-0813	DeepSeek	71.1	percent
DeepSeek V4 Flash	#5	deepseek-v4-flash	DeepSeek	68.7	percent
DeepSeek V4 Flash 0731	#6	deepseek-v4-flash-0731	DeepSeek	68.7	percent
GLM-5.2	#7	glm-5-2	Z.AI	61.8	percent
DeepSeek V4 Pro	#8	deepseek-v4-pro	DeepSeek	41.8	percent

Benchmark score table (8 models)

One best compatible score per canonical product · higher is better

DeepSeek DSBench FullStack leaderboard ranked by score
Rank	Model	Provider	License	Evidence use	Score
#1	Claude Fable 5 claude-fable-5	Anthropic	closed	Reference only	77.2%
#2	Kimi K3 kimi-k3	Moonshot AI	closed	Reference only	73.7%
#3	Claude Opus 4.8 claude-opus-4-8	Anthropic	closed	Reference only	71.6%
#4	DeepSeek V4 Pro 0813 DeepSeek-V4-Pro-0813	DeepSeek	—	Reference only	71.1%
#5	DeepSeek V4 Flash deepseek-v4-flash	DeepSeek	open	Estimated reference	68.7%
#6	DeepSeek V4 Flash 0731 deepseek/deepseek-v4-flash-0731	DeepSeek	open	Reference only	68.7%
#7	GLM-5.2 glm-5-2	Z.AI	open	Reference only	61.8%
#8	DeepSeek V4 Pro deepseek-v4-pro	DeepSeek	open	Reference only	41.8%

The top of this snapshot is led by Claude Fable 5 at 77.2%; third place is 5.6 points behind. The top-8 spread is 35.4 points.

Configuration-preserving evidencePublished source rows (9)Exact source labels, configurations, versions and dated observations. Same-snapshot canonical/configuration projections are shown once.

Every distinct dated source-published row for DeepSeek DSBench FullStack, including exact model configuration labels
Exact published model / configuration	Canonical product	Score	Version / system	Evidence	Source / observed
Fable 5 (with fallback) claude-fable-5-deepseek-0813-release-unspecified	Claude Fable 5	77.2%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
Kimi K3 kimi-k3-deepseek-0813-release-unspecified	Kimi K3	73.7%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
Claude Opus 4.8 claude-opus-4-8-deepseek-0813-release-unspecified	Claude Opus 4.8	71.6%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
DeepSeek V4 Pro 0813 deepseek-v4-pro-0813-deepseek-0813-release-unspecified	DeepSeek V4 Pro 0813	71.1%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
DeepSeek V4 Flash 0731 deepseek-v4-flash-0731-deepseek-0813-release-unspecified	DeepSeek V4 Flash 0731	68.7%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
GLM-5.2 glm-5-2-deepseek-0813-release-unspecified	GLM-5.2	61.8%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
DeepSeek V4 Pro Preview deepseek-v4-pro-deepseek-0813-release-unspecified	DeepSeek V4 Pro	41.8%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
DeepSeek V4 Flash Preview deepseek-v4-flash-deepseek-0813-release-unspecified	DeepSeek V4 Flash	37%	August 2026DeepSeek V4 Pro 0813 official GA comparison table	Reference only provider-reported	DeepSeek permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
Exact BenchLM registry variant DeepSeek V4 Flash (Max); bulk export does not retain a complete upstream harness configuration. deepseek-v4-flash-max	DeepSeek V4 Flash	68.7%	2026Source-native system	Reference only source-checked	BenchLM public datasets — 2026-08-01 ↗Observed 2026-08-01 · checked 2026-08-01

Compare top models

Claude Fable 5 vs Kimi K3 Kimi K3 vs Claude Opus 4.8 Claude Opus 4.8 vs DeepSeek V4 Pro 0813 DeepSeek V4 Pro 0813 vs DeepSeek V4 Flash

About DeepSeek DSBench FullStack

Definition and scoring

Organisation: DeepSeek-AI
Category: Coding
Version: 2026
Direction: higher is better
Ranking use: Reference
Contamination risk: Unknown

DeepSeek's internal full-stack coding-agent benchmark. Every genuine source row stays tied to its exact model label, configuration, benchmark version and evaluation system. The summary chart shows one best compatible score per canonical product; Score 2.0 admits only explicitly mapped, frozen protocols and keeps incompatible configurations separate.

Open benchmark source ↗

FAQ

Quick answers

What does DeepSeek DSBench FullStack measure?

DeepSeek's internal full-stack coding-agent benchmark.

Which model scores highest on DeepSeek DSBench FullStack?

Claude Fable 5 by Anthropic currently leads with 77.2%.

How many models are evaluated on DeepSeek DSBench FullStack?

8 canonical model products have a qualifying score, backed by 9 distinct dated source rows.

Does this affect overall Lumina rank?

No. This benchmark is display-only and does not enter the overall Lumina composite.

Coding · Benchmark profile

DeepSeek DSBench FullStack

DeepSeek's internal full-stack coding-agent benchmark.

Reference2026Active8 models

Data verified 1 Aug 2026 · Methodology 2.2.0

Benchmark score on DeepSeek DSBench FullStack