Skip to content

Benchmark

AppWorld Test-Normal Level 3: Task Goal Completion

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models and systems measured
26
Scores recorded
26
Sources
1
IQ weight
No weight published

Score distribution

35.7%
Mean
25.4%
Median
31.9%
Std dev
1.6%
10th percentile
84.1%
90th percentile
0%
Min
0%92.1%
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Claude Opus 4.6Button Agent (f072cd07-6ae0fbdf)· AppWorld Leaderboard data docs92.1%
claude-opus-4-7LARA MAS (23715209-d5c9ade3)· AppWorld Leaderboard data docs79.4%
DeepSeekCoderFullCodeRefl (0465049f-bff9fca2)· AppWorld Leaderboard data docs1.6%
DeepSeekCoderPlanExec (871bf9fd-4de66f57)· AppWorld Leaderboard data docs0%
DeepSeekCoderReAct (9ae299b8-25c255f5)· AppWorld Leaderboard data docs0%
GLM-5.3AtlasHarness (65df675b-ca5e4771)· AppWorld Leaderboard data docs87.3%
GLM-5.3-FlashIBM CUGA (a2db3f1e-2c28b1b2)· AppWorld Leaderboard data docs84.1%
GPT-4 TurboFullCodeRefl (571aa685-a8671c57)· AppWorld Leaderboard data docs1.6%
GPT-4 TurboIPFunCall (08d2ff3e-a0589430)· AppWorld Leaderboard data docs23.8%
GPT-4 TurboPlanExec (c722a232-74b9384b)· AppWorld Leaderboard data docs7.9%
GPT-4 TurboReAct (1f0055c0-123849bb)· AppWorld Leaderboard data docs1.6%
GPT-4.1IBM CUGA (8a02d49a-0df800d6)· AppWorld Leaderboard data docs54%
gpt-4.1-miniLARA MAS (695a4b55-1b68dde4)· AppWorld Leaderboard data docs42.9%
GPT-4oFullCodeRefl (d035381e-2d30c0e3)· AppWorld Leaderboard data docs12.7%
GPT-4oIPFunCall (3a889687-923696fe)· AppWorld Leaderboard data docs25.4%
GPT-4oPlanExec (ac5b742b-aa0243da)· AppWorld Leaderboard data docs20.6%
GPT-4oReAct (3414a75f-3df6a61f)· AppWorld Leaderboard data docs23.8%
GPT-4oReAct + 2 SetBSR Demos (1c12639a-57a0ac34)· AppWorld Leaderboard data docs42.9%
gpt-5.6-lunakecaipan capybara (1095b768-6513e0c6)· AppWorld Leaderboard data docs69.8%
gptoss-20bRooAGI_agentbuilder_memtune_rl (5acab4bb-badb978f)· AppWorld Leaderboard data docs50.8%
LLaMA3-70BFullCodeRefl (baf95530-cee120fc)· AppWorld Leaderboard data docs1.6%
LLaMA3-70BPlanExec (77f25849-862250c4)· AppWorld Leaderboard data docs0%
LLaMA3-70BReAct (4077db41-07bc5204)· AppWorld Leaderboard data docs7.9%
Qwen2.5-32BLOOP (e5e08b7e-d8e6474f)· AppWorld Leaderboard data docs46%
Qwen3-14BAlibaba Cloud ApsaraLab AgentRL (cabab065-0caee3c3)· AppWorld Leaderboard data docs71.4%
Qwen3.5-27BSynthAPI (3bd98253-d4fa6c8e)· AppWorld Leaderboard data docs79.4%

Data credits: AppWorld leaderboard (Stony Brook NLP). Aggregate scores from the official leaderboard; no results licence asserted. Protected bundles and experiment outputs are not used