Benchmark Almanac

PowerBench

GPT-5.6 Sol 74.2% leads since September 2026
Not in model cards

Agents retrieve equipment records and telemetry to answer 300 power-system questions. Macro Joint Accuracy averages fully correct answers equally across three task families.

The full Chinese dataset requires a request to the makers. The repository includes code and a one-equipment demo, with 50 tool calls per question.

Released
September 2026
Kind
General assistants
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Measures
Capability
Language
Chinese
In and out
Text, Time series → Actions, Text
Interaction
Agentic
Environment
APIs and tools
Items from
Real work and use
Refresh
One fixed set
Run by
A host
In views
Agents
Strongest model tested
GPT-5.6 Sol #14 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026

In the authors’ words

Large language model (LLM) agents offer new opportunities for automated analysis in industry. However, rigorous evaluation of such agents-for example, within power system scenarios-remains hindered: real operational data are confidential, and existing public resources fail to fully capture the chained dependencies and heterogeneous evidence. To address this gap, we propose PowerBench, comprising (1) a generation framework that derives interconnected heterogeneous operational data through a common dependency chain, and (2) a synthetic dataset generated by this framework. The dataset covers 761 devices across 100 device types, with 13.35 million hourly telemetry records spanning two years and 24,939 operational documents. Building on this dataset, we construct 300 questions across three task families that evaluate frontier LLMs' ability to complete analysis tasks that require autonomous evidence retrieval and reasoning across interconnected and heterogeneous data under restricted tool calls and time budgets. Results demonstrate that the evaluated frontier LLMs remain challenged on these tasks: the best model reaches only 74.2% joint accuracy. Our trace analysis further reveals that model performance varies across evidence discovery, content retrieval, tool use, reasoning over evidence, and answer submission. These findings provide detailed insights for evaluating LLM agents and guiding their reliable deployment in industry. The framework, dataset, and benchmark tasks are available at https://github.com/open-compass/PowerBench.

Abstract of PowerBench: A Benchmark for Agentic Retrieval and Reasoning in Power Systems (arXiv:2609.34492, CC0)
9
Models scored in its paper

GPT-5.6 Sol leads with 74.2%

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best.

  1. 1 GPT-5.6 Sol 50 tool calls; 300 s 74.2%
  2. 2 GLM-5.3 50 tool calls; 300 s 63.3%
  3. 3 Doubao Seed Evolving Nov 2026, 50 tool calls; 300 s 59.3%
  4. 4 DeepSeek-V4-Pro 50 tool calls; 300 s 46.5%
  5. 5 Gemini 3.6 Flash 50 tool calls; 300 s 46.4%
  6. 6 Kimi K3 50 tool calls; 300 s 45.5%
  7. 7 MiniMax-M3 50 tool calls; 300 s 25.7%
  8. 8 Qwen 3.8 Max 50 tool calls; 300 s 18.3%
  9. 9 Llama 4 Maverick 50 tool calls; 300 s 2.20%

Source: The PowerBench paper

#ModelScore ±SettingDateSource
1 GPT-5.6 Sol 74.2% 69.5 to 78.7 50 tool calls; 300 s Jul 2026 Official leaderboard
2 GLM-5.3 63.3% ±4 50 tool calls; 300 s Aug 2026 Official leaderboard
3 Doubao Seed Evolving Nov 2026 59.3% 55.7 to 62.6 50 tool calls; 300 s Nov 2026 Official leaderboard
4 DeepSeek-V4-Pro 46.5% 41.7 to 51.1 50 tool calls; 300 s Apr 2026 Official leaderboard
5 Gemini 3.6 Flash 46.4% 41.3 to 51.3 50 tool calls; 300 s Jul 2026 Official leaderboard
6 Kimi K3 45.5% ±5 50 tool calls; 300 s Jul 2026 Official leaderboard
7 MiniMax-M3 25.7% 21.3 to 30.4 50 tool calls; 300 s Jun 2026 Official leaderboard
8 Qwen 3.8 Max 18.3% 13.9 to 23 50 tool calls; 300 s Aug 2026 Official leaderboard
9 Llama 4 Maverick 2.20% 0.7 to 4.1 50 tool calls; 300 s Apr 2025 Official leaderboard

Rankings by part

We read 3 parts from the benchmark’s paper: TRR, CRR and CoRR. Each ranks models on one part of the test. Doubao Seed Evolving leads 1 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.

TRR

9 models we track have a score here.

  1. 1 Doubao Seed Evolving Nov 2026 ↑2 86.7%
  2. 2 GPT-5.6 Sol ↓1 85%
  3. 3 GLM-5.3 ↓1 80%
  4. 4 Gemini 3.6 Flash ↑1 55%
  5. 5 DeepSeek-V4-Pro ↓1 50%
  6. 6 Kimi K3 48.3%
  7. 7 MiniMax-M3 15%
  8. 8 Qwen 3.8 Max 15%
  9. 9 Llama 4 Maverick 0%

Number one over time

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

OpenAI
Benchmark paper 1 leader
GPT-5.6 Sol 74.2%

Source: The PowerBench paper

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
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-5.6 Sol DeepSeek-V4-Pro Llama 4 Maverick 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-5.6 Sol DeepSeek-V4-Pro

Source: Official PowerBench 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 benchmark’s paper (9 entries).
  • Pass Models scored 9 models we track.
  • Pass Newest score The newest is from November 2026, for Doubao Seed Evolving.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.6 Sol (#14).
  • Pass Headroom The best is 74.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 9 rows for models we track at the last read.
  • Pass Last read November 20th, 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