AppWorld evaluates agents completing everyday tasks across nine simulated apps through code and APIs. Tasks include coordinating messages, shopping, and personal information. Scores here measure Task Goal Completion on the Test-Challenge split: the percentage of tasks for which all checks pass. Agent methods are retained with each result.
| Model / creator | Version | Reported score |
|---|---|---|
 Claude Opus 4.7Anthropic · Proprietary | Test-Challenge | 85.6% |
 GPT-5.6 LunaOpenAI · Proprietary | Test-Challenge | 73.4% |
| Claude Opus 4.6Anthropic · Proprietary | Test-Challenge | 72.4% |
 Qwen3-14BQwen · Open source / open weights | Test-Challenge | 67.6% |
| GPT-4.1OpenAI · Proprietary | Test-Challenge | 57.6% |
| Qwen2.5-32BQwen · Open source / open weights | Test-Challenge | 47.2% |
| GPT-4oOpenAI · Proprietary | Test-Challenge | 38.9% |
| GPT-4.1 miniOpenAI · Proprietary | Test-Challenge | 37.6% |
| GPT-4 TurboOpenAI · Proprietary | Test-Challenge | 17.5% |
| gpt-oss-20bOpenAI · Open source / open weights | Test-Challenge | 15.3% |
 LLaMA3-70BMeta · Open source / open weights | Test-Challenge | 7% |
 DeepSeekCoderDeepSeek · Open source / open weights | Test-Challenge | 5.8% |
agent: LARA MAS · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: claude-opus-4-7 · split: Test-Challenge · test challenge sgc: 77 · test normal sgc: 82.1 · test normal tgc: 88.7
AppWorld research team ↗agent: LARA MAS · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: gpt-4.1-mini · split: Test-Challenge · test challenge sgc: 20.1 · test normal sgc: 50 · test normal tgc: 61.9
AppWorld research team ↗agent: kecaipan capybara · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: gpt-5.6-luna · split: Test-Challenge · test challenge sgc: 52.5 · test normal sgc: 73.2 · test normal tgc: 85.1
AppWorld research team ↗agent: Button Agent · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: Claude Opus 4.6 · split: Test-Challenge · test challenge sgc: 64.8 · test normal sgc: 89.3 · test normal tgc: 91.1
AppWorld research team ↗agent: RooAGI_agentbuilder_memtune_rl · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: gptoss-20b · split: Test-Challenge · test challenge sgc: 7.9 · test normal sgc: 58.9 · test normal tgc: 76.2
AppWorld research team ↗agent: Alibaba Cloud ApsaraLab AgentRL · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: Qwen3-14B · split: Test-Challenge · test challenge sgc: 50.4 · test normal sgc: 80.4 · test normal tgc: 86.9
AppWorld research team ↗agent: ReAct + 2 SetBSR Demos · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4o · split: Test-Challenge · test challenge sgc: 23 · test normal sgc: 57.1 · test normal tgc: 68.5
AppWorld research team ↗agent: IBM CUGA · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4.1 · split: Test-Challenge · test challenge sgc: 48.2 · test normal sgc: 62.5 · test normal tgc: 73.2
AppWorld research team ↗agent: LOOP · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: Qwen2.5-32B · split: Test-Challenge · test challenge sgc: 28.8 · test normal sgc: 53.6 · test normal tgc: 72.6
AppWorld research team ↗agent: ReAct · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4o · split: Test-Challenge · test challenge sgc: 13 · test normal sgc: 32.1 · test normal tgc: 48.8
AppWorld research team ↗agent: PlanExec · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4o · split: Test-Challenge · test challenge sgc: 7.9 · test normal sgc: 23.2 · test normal tgc: 44.6
AppWorld research team ↗agent: FullCodeRefl · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4o · split: Test-Challenge · test challenge sgc: 12.2 · test normal sgc: 26.8 · test normal tgc: 33.9
AppWorld research team ↗agent: IPFunCall · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4o · split: Test-Challenge · test challenge sgc: 10.1 · test normal sgc: 16.1 · test normal tgc: 32.1
AppWorld research team ↗agent: ReAct · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4 Turbo · split: Test-Challenge · test challenge sgc: 5.8 · test normal sgc: 12.5 · test normal tgc: 26.8
AppWorld research team ↗agent: IPFunCall · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4 Turbo · split: Test-Challenge · test challenge sgc: 9.3 · test normal sgc: 21.4 · test normal tgc: 30.4
AppWorld research team ↗agent: FullCodeRefl · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4 Turbo · split: Test-Challenge · test challenge sgc: 7.2 · test normal sgc: 19.6 · test normal tgc: 25.6
AppWorld research team ↗agent: PlanExec · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: GPT-4 Turbo · split: Test-Challenge · test challenge sgc: 3.6 · test normal sgc: 16.1 · test normal tgc: 32.7
AppWorld research team ↗agent: FullCodeRefl · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: LLaMA3-70B · split: Test-Challenge · test challenge sgc: 4.3 · test normal sgc: 17.9 · test normal tgc: 24.4
AppWorld research team ↗agent: FullCodeRefl · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: DeepSeekCoder · split: Test-Challenge · test challenge sgc: 2.9 · test normal sgc: 8.9 · test normal tgc: 13.1
AppWorld research team ↗agent: ReAct · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: LLaMA3-70B · split: Test-Challenge · test challenge sgc: 0 · test normal sgc: 8.9 · test normal tgc: 20.8
AppWorld research team ↗agent: ReAct · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: DeepSeekCoder · split: Test-Challenge · test challenge sgc: 0.7 · test normal sgc: 1.8 · test normal tgc: 7.1
AppWorld research team ↗agent: PlanExec · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: LLaMA3-70B · split: Test-Challenge · test challenge sgc: 0.7 · test normal sgc: 1.8 · test normal tgc: 8.9
AppWorld research team ↗agent: PlanExec · metric: TGC · selection: All 23 leaderboard submissions; latest report date and best score at that date shown per model. Normal-split and SGC values retained in the source data/settings, not mixed into the ranking. · source model label: DeepSeekCoder · split: Test-Challenge · test challenge sgc: 0 · test normal sgc: 0 · test normal tgc: 1.8
AppWorld research team ↗SWE-bench evaluates software engineering agents on resolving real issues in existing code repositories.
GDPval-AAGDPval-AA evaluates models on professional knowledge-work tasks from OpenAI’s GDPval dataset, including creating documents, spreadsheets, slides, and other work products.
GPQA DiamondGPQA Diamond tests graduate-level scientific knowledge and reasoning with difficult multiple-choice questions in biology, chemistry, and physics.
