Benchmark Almanac

MobilePA-Bench

Qwen-Planner-Agent 77% tops its leaderboard GPT-6 Astra 76.9% leads the models we track
Not in model cards

Agents complete 1,705 mobile planning tasks using tools, persistent memory, reusable skills, and specialist agents. Task-specific checks evaluate tool calls, state changes, and interaction behavior. Additional gates check required memory access and skill loading.

The simulated environment excludes visual perception and device execution. The current protocol produces 1,905 scored trajectories per system. Queries and gold labels stay on the maker's evaluation server. Behavioral checks use two fixed model judges.

Released
August 2026
Kind
General assistants
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text → Text, Actions
Interaction
Multi-agent
Items from
Generated
Refresh
One fixed set
Run by
A host
In views
Agents, Robotics
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
October 2026
Scores refreshed
November 18th, 2026

In the authors’ words

Mobile planning requires a central agent to coordinate tools, persistent user context, reusable procedures, and specialized sub-agents while adapting to execution feedback. Evaluating these workflows requires distinguishing task completion from compliance with prescribed resource and procedural requirements. We introduce MobilePA-Bench, a benchmark comprising 1,705 tasks across eight query task domains and 212 mobile tools in a stateful simulated environment. It combines three features: (1) unified capability coverage, profiling Tool Use, Memory Usage, Skill Usage, and Sub-agent Collaboration under a common mobile protocol; (2) evidence-aligned verification, matching each task to a fixed tool-call, state-change, or interaction-behavior checker rather than imposing one completion criterion on all tasks; and (3) capability-gated evaluation, requiring task completion alongside prescribed memory checks and, in skill-only routing, complete route loading before business-tool use. With MobilePA-Bench, we systematically evaluate open- and closed-weight LLMs, revealing capability-specific weaknesses in complex mobile task execution. Fine-grained analyses distinguish failures in resource retrieval and procedural compliance from downstream execution errors, providing diagnostic insights beyond aggregate task-success rates. We further train a Qwen3.8-27B baseline with online reinforcement learning in the sandbox, whose improved performance provides initial evidence of the environment's utility for planner development.

Abstract of MobilePA-Bench: Benchmarking Mobile Planner Agents on Complex Real-World Tasks (arXiv:2608.23035, CC0)
21
Entries on its own leaderboard, 20 models we track

Qwen-Planner-Agent tops the leaderboard with 77% GPT-6 Astra leads with 76.9%

Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-6 Astra, at 76.9%. This view shows only models we track, each at its best.

  1. 1 Qwen-Planner-Agent 77%
  2. 2 GPT-6 Astra 76.9%
  3. 3 Claude Opus 5 75.7%
  4. 4 Claude Fable 5 74.5%
  5. 5 GLM-5.3 73.9%
  6. 6 DeepSeek V4.1 Flash 1017 72%
  7. 7 Qwen 3.8 Max 71.8%
  8. 8 Muse Spark 1.3 71.5%
  9. 9 Qwen3.8 Max (0902) 71.4%
  10. 10 Kimi K3 69.6%
  11. 11 Gemini 3.6 Flash 69.6%
  12. 12 GLM-5.3-Flash 68.3%
  13. 13 Gemini 3.1 Pro 68.2%
  14. 14 Qwen 3.8 27B 67.2%
  15. 15 GLM-5.2 66.1%
  16. 16 Claude Opus 4.8 66%
  17. 17 Qwen3.7-Max 64.9%
  18. 18 Seed2.1 Pro 64.9%
  19. 19 Qwen3.7-Plus 62.5%
  20. 20 GPT-5.5 61.9%
  21. 21 GPT-5.6 Sol 60.2%

Source: Official MobilePA-Bench leaderboard

#EntryScoreSettingDateSource
1 Qwen-Planner-Agent 77% Official leaderboard
2 GPT-6 Astra 76.9% Sep 2026 Official leaderboard
3 Claude Opus 5 75.7% Jul 2026 Official leaderboard
4 Claude Fable 5 74.5% Jun 2026 Official leaderboard
5 GLM-5.3 73.9% Aug 2026 Official leaderboard
6 DeepSeek V4.1 Flash 1017 72% Oct 2026 Official leaderboard
7 Qwen 3.8 Max 71.8% Aug 2026 Official leaderboard
8 Muse Spark 1.3 71.5% Sep 2026 Official leaderboard
9 Qwen3.8 Max (0902) 71.4% Sep 2026 Official leaderboard
10 Kimi K3 69.6% Jul 2026 Official leaderboard
11 Gemini 3.6 Flash 69.6% Jul 2026 Official leaderboard
12 GLM-5.3-Flash 68.3% Aug 2026 Official leaderboard
13 Gemini 3.1 Pro 68.2% Feb 2026 Official leaderboard
14 Qwen 3.8 27B 67.2% Aug 2026 Official leaderboard
15 GLM-5.2 66.1% Jun 2026 Official leaderboard
16 Claude Opus 4.8 66% May 2026 Official leaderboard
17 Qwen3.7-Max 64.9% May 2026 Official leaderboard
18 Seed2.1 Pro 64.9% Jun 2026 Official leaderboard
19 Qwen3.7-Plus 62.5% Jun 2026 Official leaderboard
20 GPT-5.5 61.9% Apr 2026 Official leaderboard
21 GPT-5.6 Sol 60.2% Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6 Astra 76.9% Sep 2026 Official leaderboard
2 Claude Opus 5 75.7% Jul 2026 Official leaderboard
3 Claude Fable 5 74.5% Jun 2026 Official leaderboard
4 GLM-5.3 73.9% Aug 2026 Official leaderboard
5 DeepSeek V4.1 Flash 1017 72% Oct 2026 Official leaderboard
6 Qwen 3.8 Max 71.8% Aug 2026 Official leaderboard
7 Muse Spark 1.3 71.5% Sep 2026 Official leaderboard
8 Qwen3.8 Max (0902) 71.4% Sep 2026 Official leaderboard
9 Kimi K3 69.6% Jul 2026 Official leaderboard
10 Gemini 3.6 Flash 69.6% Jul 2026 Official leaderboard
11 GLM-5.3-Flash 68.3% Aug 2026 Official leaderboard
12 Gemini 3.1 Pro 68.2% Feb 2026 Official leaderboard
13 Qwen 3.8 27B 67.2% Aug 2026 Official leaderboard
14 GLM-5.2 66.1% Jun 2026 Official leaderboard
15 Claude Opus 4.8 66% May 2026 Official leaderboard
16 Qwen3.7-Max 64.9% May 2026 Official leaderboard
17 Seed2.1 Pro 64.9% Jun 2026 Official leaderboard
18 Qwen3.7-Plus 62.5% Jun 2026 Official leaderboard
19 GPT-5.5 61.9% Apr 2026 Official leaderboard
20 GPT-5.6 Sol 60.2% Jul 2026 Official leaderboard

Rankings by part

We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Tool use, Memory and Skills. Claude Fable 5 leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Tool use

20 models we track have a score here.

  1. 1 Claude Opus 5 ↑1 77.6%
  2. 2 Claude Fable 5 ↑1 77.2%
  3. 3 GLM-5.3 ↑1 76.4%
  4. 4 GPT-6 Astra ↓3 75.8%
  5. 5 Gemini 3.1 Pro ↑7 75.4%
  6. 6 Gemini 3.6 Flash ↑4 74.7%
  7. 7 GLM-5.2 ↑7 73.8%
  8. 8 DeepSeek V4.1 Flash 1017 ↓3 73.8%
  9. 9 Qwen 3.8 Max ↓3 73.7%
  10. 10 Muse Spark 1.3 ↓3 73.1%

2 models have held number one on the official leaderboard since its release in August 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic OpenAI
Own leaderboard 2 leaders
GPT-6 Astra 76.9%

Source: Official MobilePA-Bench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Claude Fable 5 Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Gemini 3.1 Pro

Source: Official MobilePA-Bench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Tool use

All 265 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Neon 56.3%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Gemini 4 Argon 90.7%
  • BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (21 entries).
  • Pass Models scored 20 models we track, plus 1 other entry on its leaderboard.
  • Pass Newest score The newest is from October 2026, for DeepSeek V4.1 Flash 1017.
  • Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
  • Pass Headroom The best is 76.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on tongyi-mai.github.io: 20 rows for models we track and 1 other entry at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check