Suite results
Suite Score Avg / 20 Tests n tok/s Run
▶ General capability (13-task real-workload suite) 1a 176.8 / 260 13.6 13/13 n=1.1 144.3 run 2026-08-04
| Test | Category | Score | | tok/s |
| A1 | Dev/Ops Scripting | 3/20 | | 105.9 |
| A2 | Dev/Ops Scripting | 12/20 | | 158.6 |
| A3 | Dev/Ops Scripting | 9/20 | | 121.4 |
| A4 | Dev/Ops Scripting | 6/20 | | 141.2 |
| A5 | Dev/Ops Scripting | 15/20 | | 135.3 |
| A6 | Dev/Ops Scripting | 13/20 | | 141.6 |
| B1 | Document Processing | 20/20 | | 148.6 |
| B2 | Document Processing | 20/20 | | 153.8 |
| B3 | Document Processing | 8/20 | | 99.2 |
| C1 | Content Production | 17/20 | | 145.3 |
| C2 | Content Production | 18/20 | | 143.4 |
| C3 | Content Production | 18/20 | | 145.1 |
| C4 | Content Production | 17/20 | | 148.9 |
Category mean Dev/Ops Scripting 9.7Document Processing 16Content Production 17.5
▶ Agentic tool-calling & protocol adherence 1b 148.8 / 160 18.6 8/8 n=1 177.8 run 2026-09-21
| Test | Category | Score | | tok/s |
| WA1 | Tool Calling & Protocol Adherence | 20/20 | | 88 |
| WA2 | Tool Calling & Protocol Adherence | 20/20 | | 86.6 |
| WA3 | Tool Calling & Protocol Adherence | 20/20 | | 104.3 |
| WA4 | Tool Calling & Protocol Adherence | 20/20 | | 79.1 |
| WA5 | Tool Calling & Protocol Adherence | 20/20 | | 143.3 |
| WB1 | Agent Robustness & State Management | 18/20 | | 165 |
| WB2 | Agent Robustness & State Management | 20/20 | | 167.3 |
| WB3 | Agent Robustness & State Management | 11/20 | | 158.9 |
Category mean Tool Calling & Protocol Adherence 20Agent Robustness & State Management 16.3
▶ Coding depth 1c 100.8 / 180 11.2 9/9 n=1 155.4 run 2026-09-30
| Test | Category | Score | | tok/s |
| K1 | Coding — Concurrency | 16/20 | | 149.5 |
| K2 | Coding — Algorithms | 13/20 | | 166.7 |
| K3 | Coding — Security Review | 9/20 | | 167 |
| K4 | Coding — Performance | 16/20 | | 162 |
| K5 | Coding — Refactoring Under Constraints | 15/20 | | 154.6 |
| K6 | Coding — Test Writing | 16/20 | | 162.5 |
| K7 | Coding — React Component (typed) | 6/20 | | 158.4 |
| K8 | Coding — TypeScript Type System | 9/20 | | 153.4 |
| K9 | Coding — React Debugging | 1/20 | | 81.2 |
Category mean Coding — Concurrency 16Coding — Algorithms 13Coding — Security Review 9Coding — Performance 16Coding — Refactoring Under Constraints 15Coding — Test Writing 16Coding — React Component (typed) 6Coding — TypeScript Type System 9Coding — React Debugging 1
▶ Content-production depth 1e 94.8 / 120 15.8 6/6 n=1 168 run 2026-09-25
| Test | Category | Score | | tok/s |
| E1 | SOW & Requirements Decomposition | 20/20 | | 160.8 |
| E2 | SOW & Requirements Decomposition | 20/20 | | 154.7 |
| E3 | Executive & Proposal Content | 17/20 | | 152.6 |
| E4 | Executive & Proposal Content | 15/20 | | 166 |
| E5 | Executive & Proposal Content | 15/20 | | 155.5 |
| E6 | Executive & Proposal Content | 8/20 | | 156.5 |
Category mean SOW & Requirements Decomposition 20Executive & Proposal Content 13.8
▶ Long-context retrieval & synthesis 1g 120 / 120 20 6/6 n=1 73.8 run 2026-09-23
| Test | Category | Score | | tok/s |
| L1 | Long-Context — Needle Retrieval | 20/20 | | 20.9 |
| L2 | Long-Context — Grounded QA with Distractors | 20/20 | | 7.2 |
| L3 | Long-Context — Ops Log Reasoning | 20/20 | | 15.6 |
| L4 | Long-Context — Faithful Summarization | 20/20 | | 32.9 |
| L5 | Long-Context — Instruction Retention | 20/20 | | 7.5 |
| L6 | Long-Context — Full-Haystack QA | 20/20 | | 1.8 |
Category mean Long-Context — Needle Retrieval 20Long-Context — Grounded QA with Distractors 20Long-Context — Ops Log Reasoning 20Long-Context — Faithful Summarization 20Long-Context — Instruction Retention 20Long-Context — Full-Haystack QA 20
▶ Long-context multi-needle (MRCR) 1g2 24 / 60 8 3/3 n=1 48.8 run 2026-09-19
| Test | Category | Score | | tok/s |
| L7 | Long-Context — Multi-Needle Retrieval (MRCR) | 10/20 | | 30.2 |
| L8 | Long-Context — Multi-Needle Retrieval (MRCR) | 10/20 | | 11.6 |
| L9 | Long-Context — Multi-Needle Retrieval (MRCR) | 4/20 | | 5 |
▶ Live one-shot builds (runtime-verified) 1h 32.8 / 80 8.2 4/4 n=1 139 run 2026-09-19
| Test | Category | Score | | tok/s |
| H1 | Web-Estate One-Shot Builds | 12/20 | | 119.3 |
| H2 | Web-Estate One-Shot Builds | 12/20 | | 143.1 |
| H3 | Web-Estate One-Shot Builds | 7/20 | | 143.7 |
| H4 | Web-Estate One-Shot Builds | 2/20 | | 142.9 |
▶ Brand-constrained static build (negative-constraint adherence) 1j 51 / 100 10.2 5/5 n=1 161.4 run 2026-09-22
| Test | Category | Score | | tok/s |
| J1 | Brand-Constrained Static Build | 10/20 | | 153.4 |
| J3 | Brand-Constrained Static Build | 8/20 | | 155.9 |
| J4 | Brand-Constrained Static Build | 12/20 | | 147.6 |
| J5 | Brand-Constrained Static Build | 13/20 | | 157 |
| J6 | Brand-Constrained Static Build | 8/20 | | 154.9 |
Each test is LLM-judged 0–20 against a fixed rubric; suite max = tests × 20.
An ADJ chip marks a cell carrying a subjective
adjustment from the maintainer; the judged score is the one shown, is unchanged,
and remains what the axes, ranks and tie bands are computed from. Hover for the reason.
Rows with a ▶ expand to the per-test breakdown.
Where fewer tests ran than the suite total, unrun tests count as zero toward the axis score.
See methodology.