APEX-Agents
Integrity rank #31 of 65 · 49 models scored · top score 82.4 · Gemini 4 Argon
Long-horizon, cross-application agent execution on professional-services tasks (investment banking, consulting, legal work). 480 tasks across 33 simulated 'worlds' (160 per domain: banking/consulting/legal).
Strongest on discrimination, weakest on contamination resistance. Caveat: the public test set does not rule out training-data contamination.
Reported scores — protocols may differ. How we compare
| # | Model | Score | Evidence | Measured |
|---|---|---|---|---|
| 1 | Gemini 4 Argon | 82.4 | unverifiedT2 | 2026-09-30 |
| 2 | Claude Sonnet 5.5 | 75.5 | unverifiedT2 | 2026-09-28 |
| 3 | Claude Opus 5.5 | 73.5 | unverifiedT2 | 2026-09-22 |
| 4 | Claude Fable 5.1 | 68.6 | unverifiedT2 | 2026-09-01 |
| 5 | Gemini 3.7 Flash | 67.8 | unverifiedT2 | 2026-08-13 |
| 6 | Claude Opus 5 | 65.8 | unverifiedT2 | 2026-07-24 |
| 7 | Grok 4.6 | 65.3 | unverifiedT2 | 2026-08-12 |
| 8 | GPT-6 Astra | 64.7 | unverifiedT2 | 2026-09-03 |
| 9 | Gemini 3.8 Flash | 64.3 | unverifiedT2 | 2026-09-02 |
| 10 | Claude Fable 5 | 63.6 | unverifiedT2 | 2026-06-09 |
| 11 | Qwen 3.8 Max | 63.3 | unverifiedT2 | 2026-08-02 |
| 12 | GPT-6.1 Sol | 60 | unverifiedT2 | 2026-09-29 |
| 13 | GPT-5.6 Terra | 58.2 | unverifiedT2 | 2026-07-09 |
| 14 | Muse Spark 1.3 | 57.8 | unverifiedT2 | 2026-09-02 |
| 15 | GLM-5.3 | 56.6 | unverifiedT2 | 2026-08-14 |
| 16 | Grok 4.5 | 56.2 | unverifiedT2 | 2026-07-08 |
| 17 | Claude Haiku 5.5 | 56.1 | unverifiedT2 | 2026-10-07 |
| 18 | GPT-5.5 | 55.1 | unverifiedT2 | 2026-04-23 |
| 19 | Grok 4.7 | 54.6 | unverifiedT2 | 2026-09-21 |
| 20 | Claude Sonnet 5 | 54.5 | unverifiedT2 | 2026-06-30 |
| 21 | GPT-6 Sol | 54.3 | unverifiedT2 | 2026-09-22 |
| 22 | GLM-5.3-Flash | 52.8 | unverifiedT2 | 2026-08-20 |
| 23 | GPT-5.4 | 52.4 | unverifiedT2 | 2026-03-05 |
| 24 | GPT-5.6 Sol | 51.4 | unverifiedT2 | 2026-07-09 |
| 25 | Kimi K3 | 50.6 | unverifiedT2 | 2026-07-16 |
| 26 | Claude Opus 4.7 | 49.2 | unverifiedT2 | 2026-04-16 |
| 27 | Claude Opus 4.8 | 48.9 | unverifiedT2 | 2026-05-28 |
| 28 | Qwen 3.8 27B | 47.5 | unverifiedT2 | 2026-08-14 |
| 29 | DeepSeek V4 Pro 0813 | 47.3 | unverifiedT2 | 2026-08-13 |
| 30 | Gemini 3.6 Flash | 46.9 | unverifiedT2 | 2026-07-21 |
| 31 | Claude Opus 4.6 | 46.3 | unverifiedT2 | 2026-02-05 |
| 32 | GLM-5.2 | 45.2 | unverifiedT2 | 2026-06-16 |
| 33 | GPT-6 Luna | 44.3 | unverifiedT2 | 2026-09-22 |
| 34 | Claude Sonnet 4.6 | 43 | unverifiedT2 | 2026-02-17 |
| 35 | GPT-5.6 Luna | 43 | unverifiedT2 | 2026-07-09 |
| 36 | GLM-5.1 | 40.9 | unverifiedT2 | 2026-04-07 |
| 37 | DeepSeek V4.1 Flash | 39.5 | unverifiedT2 | 2026-09-09 |
| 38 | MiniMax-M3 | 37.7 | unverifiedT2 | 2026-06-01 |
| 39 | Kimi K2.7 Code | 37.6 | unverifiedT2 | 2026-06-12 |
| 40 | Muse Spark 1.2 | 36.4 | unverifiedT2 | 2026-08-05 |
| 41 | Gemini 3.1 Pro | 35.3 | unverifiedT2 | 2026-02-19 |
| 42 | Inkling | 33.8 | unverifiedT2 | 2026-07-15 |
| 43 | Muse Spark 1.1 | 31.8 | unverifiedT2 | 2026-07-09 |
| 44 | Gemini 3.5 Flash-Lite | 29.3 | unverifiedT2 | 2026-07-21 |
| 45 | Gemini 3.5 Flash | 27.5 | unverifiedT2 | 2026-05-19 |
| 46 | Qwen3.5 397B-A17B | 24.9 | unverifiedT2 | 2026-02-13 |
| 47 | Nemotron 3 Ultra | 22.7 | unverifiedT2 | 2026-06-04 |
| 48 | DeepSeek-V3.2 | 21.3 | unverifiedT2 | 2025-12-01 |
| 49 | gpt-oss-120b | 4.4 | unverifiedT2 | 2025-08-05 |
Continue from this leaderboard into the tightest head-to-head reads.
Weighted composite. Each dimension opens to the source-backed sub-signals it is computed from.
Claims drawn from cited facts, not live model generation.
This benchmark tests long-running agent tasks across applications in investment banking, consulting and legal work.
1 cited fact
This benchmark has scores from 49 tracked models, with a recorded score range spanning 78.0.
2 cited facts
Its Benchmark Integrity Index score is 85, with grade A and rank 31 out of 65 tracked benchmarks.
4 cited facts
It is not flagged as saturated. The top recorded score is 82.4, with 17.6 points of headroom. The computed top-score cluster contains 1 models.
4 cited facts
The harness classification is consistent harness. Test-set privacy is recorded as unknown. The recorded contamination-window ratio is 0.959.
3 cited facts
APEX-Agents: Long-horizon, cross-application agent execution on professional-services tasks (investment banking, consulting, legal work). 480 tasks across 33 simulated 'worlds' (160 per domain: banking/consulting/legal).
Gemini 4 Argon leads APEX-Agents at 82.4. The full leaderboard above lists every recorded measurement, not just the headline number.
49 models have recorded scores on APEX-Agents, spanning a score spread of 78.
tensor.news grades APEX-Agents A for integrity (score 85/100), ranking #31 of 65 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — APEX-Agents still has headroom and continues to discriminate between models rather than bunching them at the ceiling.
Its test set is unknown, so contamination risk is on the table: a high score may partly reflect training-data overlap rather than capability.
Scores are reported under a consistent harness, so comparisons on APEX-Agents are reasonably apples-to-apples.
Every APEX-Agents measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites arxiv.org/abs/2601.14242.