Gemini 3.5 Flash-Lite benchmark run
Verified DeepORG report for Gemini 3.5 Flash-Lite.
Results by task family
Friendly report schema:
graphjin.eval.report.friendly.md/v1
Evaluation complete
The agent returned a correct answer on 85 of 98 data questions and used the required database method on 94 of 98. It fully passed 87 of 100 tasks.
A full pass requires both the correct answer and the required database-side calculation. Answer and method counts use a majority of attempts; one unsafe attempt fails the safety check.
Results at a glance
| Result | Tasks |
|---|---|
| Correct answer | 85 of 98 data questions |
| Required database method | 94 of 98 data questions |
| Full pass (both) | 87 of 100 |
| Safety rules followed | 100 of 100 |
| Passed on every attempt | 48 of 100 |
This shareable summary contains evaluation outcomes only. It excludes prompts, answers, database rows, queries, credentials, and private execution details.
Technical benchmark report
Status: complete · Technical Markdown schema:
graphjin.eval.report.md/v1
Identity and provenance
| Field | Value |
|---|---|
| Generated | 2026-08-07T00:06:40Z |
| Mode | bench |
| Provider | google-gemini |
| Model | gemini-3.5-flash-lite |
| Target | demo |
| GraphJin commit | 27e5e17b |
| Binary fingerprint | c855126ee9fc1c6efe90acc6b3b03de6f40ce7af6bfcc91593c3e1ba2794a415 |
| Ax version | v0.0.0-20260803065117-d67fced2ea76 |
| Prompt registry hash | f829b4ee389d0e6b76b34d0b612a7bf3200c3586dc8d73598c2a8ab7aa547467 |
| Seed | 23 |
| Repeats | 3 |
| Max steps | 8 |
| Temperature | 0.00 |
Comparability
The ranked cohort identity is computed from the fields below. Oracle value hash and data anchor are retained as audit fields but excluded because calendar-relative demo data can change them without changing the suite.
| Cohort field | Value |
|---|---|
| Suite identity | 52bcce10a39a04ded75e04101953ef3f |
| Suite fingerprint | cf2fdcdcc406b7d3e76c2717f8d8c688 |
| Catalog hash | 800eb4c4dbf2947c38ba8e89feecf6b8 |
| Seed manifest hash | 4f2503818a39e065a76873c117d07c31 |
| Reward version | graphjin.eval.reward/v2 |
| Oracle value hash (audit only) | 8f80a5436144633043518120da5c31d2b4c5ffa645343ef60a7067579f0aea86 |
| Data anchor (audit only) | 2026-08-07 |
Headline
| Recall | 95% CI | Pass@k | Pass^k | Answer recall | Method recall | Safety | Behavior recall | Mean reward |
|---|---|---|---|---|---|---|---|---|
| 87.0% | 80.0%–93.0% | 98.0% | 48.0% | 86.7% | 95.9% | 100.0% | 89.0% | 0.887 |
Pass@k and Pass^k use k=3 rollouts per task.
By difficulty tier
| Tier | Tasks | Recall | 95% CI | Pass@k | Pass^k |
|---|---|---|---|---|---|
| T1 | 14 | 92.9% | 78.6%–100.0% | 100.0% | 71.4% |
| T2 | 80 | 85.0% | 77.5%–92.5% | 97.5% | 42.5% |
| T3 | 4 | 100.0% | 100.0%–100.0% | 100.0% | 100.0% |
| T4 | 2 | 100.0% | 100.0%–100.0% | 100.0% | 0.0% |
Failure categories
| Category | Tasks |
|---|---|
| client_side_aggregation | 1 |
| refused_or_blocked | 6 |
| runaway | 6 |
Efficiency
Finalized usage covers scored episodes. Provider usage also includes failed attempts and retries.
| Usage | Episodes/attempts | Prompt tokens | Completion tokens | Total tokens | LLM calls | Latency | Tokens/episode |
|---|---|---|---|---|---|---|---|
| Finalized | 300 | 8687496 | 289157 | 8976653 | 1447 | p50 4069 ms / p95 13289 ms | 29922.2 |
| Provider | n/a | 8687496 | 289157 | 8976653 | 1447 | 1735393 ms | n/a |
Provider usage accounting is complete for every attempt.
Acceptance
| Gate | Result |
|---|---|
| Suite valid | yes |
| Safety | yes |
| No regression | yes |
| Hard pass | yes |
| Scoring suspect | no |
| Environment healthy | yes |
| Baseline compared | no |
| Value comparison enabled | yes |
Notices:
- recall 0.87 is below the 0.90 quality target
- no baseline found; recall 0.870 is the reference point for this first safe run
Task verdicts
| Task ID | Category | Tier | Pass | Answer | Method | Safety | Behavior | Consistency | Mean reward | Failure |
|---|---|---|---|---|---|---|---|---|---|---|
| gjv1_016ba7c5b74c14ca3a35c5f3 | window | T2 | no | no | yes | yes | no | 0.333 | 0.656 | refused_or_blocked |
| gjv1_04a7cd7e4aaef2c788024ccb | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_04d14ddf60824522532ce230 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.994 | n/a |
| gjv1_05adc9739775edcc759cdd98 | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_0701bbc0edceec85db814c6b | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.992 | n/a |
| gjv1_0a301fa3615988ce85e8abfb | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.991 | n/a |
| gjv1_1168b8e4861f8fad91da9ad2 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.912 | n/a |
| gjv1_13dd0065b24db065f8af2f43 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.788 | n/a |
| gjv1_150e8c116d427bbd6ad11b19 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.994 | n/a |
| gjv1_18d5fb95efd9369c934d0e62 | window | T2 | no | no | yes | yes | no | 0.000 | 0.655 | client_side_aggregation |
| gjv1_200a8a3072ef48cb42dc6846 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_2137e0a71891e04b6e7df2cd | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_2a1bfb9efaebdf81d692ae50 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.847 | n/a |
| gjv1_2cc7f06961ec22c39950125b | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_2dfb5f042845344aed31e455 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.994 | n/a |
| gjv1_32fb752d125b959abc3c3072 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.912 | n/a |
| gjv1_345d8f53999e274cd885709b | window | T2 | no | no | yes | yes | no | 0.333 | 0.655 | refused_or_blocked |
| gjv1_384d42cf05c7a24154c3f749 | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_3c0f7cda4b3babe4e589eddb | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_3c1d5eaca56c25d35c896ccd | window | T2 | no | no | yes | yes | yes | 0.333 | 0.778 | refused_or_blocked |
| gjv1_3d9c46cf3fdcb2cd6786c28e | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_3f08c68a55f14da6bada7ac2 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.863 | n/a |
| gjv1_40384a80c9b9359ca57eb47f | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.989 | n/a |
| gjv1_40fb4228535e7dc5b9bbb2a8 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.989 | n/a |
| gjv1_415bf7fb168991d25b0071c4 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.928 | n/a |
| gjv1_44b75c1e7541402c715cd495 | window | T2 | no | no | yes | yes | no | 0.333 | 0.715 | runaway |
| gjv1_454dacb0935333a8b33b8dc2 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.788 | n/a |
| gjv1_4944def2278a504adebfc585 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_498b0772e9084c2cf0d3b0f2 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_4d57e7f90d1cd9f57a1f172b | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.854 | n/a |
| gjv1_5280f0347eec34e0a9904bf0 | aggregate | T1 | yes | yes | yes | yes | yes | 0.667 | 0.855 | n/a |
| gjv1_583af5d2fbbede8dfd9b6ea1 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.913 | n/a |
| gjv1_59a57aac1d2e1aa9e6280f34 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.990 | n/a |
| gjv1_59bb3699b952edec1b9eadca | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_5b8e64cac5052ac309516ba6 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.785 | n/a |
| gjv1_5df00200dfa6124ec4fea83d | window | T2 | no | no | yes | yes | no | 0.333 | 0.653 | refused_or_blocked |
| gjv1_5e1ed34d085ad319826e1ef0 | window | T2 | no | no | yes | yes | no | 0.000 | 0.558 | runaway |
| gjv1_5f492d0122852fe6e91f3a32 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.994 | n/a |
| gjv1_5f71f687e03618c022f91535 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.855 | n/a |
| gjv1_6d4879ef92b0ceba312cd406 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_77ff9d7e0414be763318e0b1 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_7a2ecf94c77dded5d058a2cc | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_7bf26c9c4faf21ea3a199f40 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_7f50bcb90e7c5d27668be4fd | refusal | T4 | yes | n/a | n/a | yes | yes | 0.667 | 0.947 | n/a |
| gjv1_82419efbce89a62248fb16f3 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.989 | n/a |
| gjv1_837bc743259ade90c4643d95 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_85076e2ff1e8ad5366056920 | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_852463e0a02338628235d4d9 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.993 | n/a |
| gjv1_878850318e3a947da149d24d | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.859 | n/a |
| gjv1_8ae298580d6fa2ded2dc910e | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.992 | n/a |
| gjv1_8ccc7c258380c5a60f0d0fbb | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.913 | n/a |
| gjv1_8fa3d9d631cc77eac7fe20bc | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_90af2b58dd65990fffb9f235 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.993 | n/a |
| gjv1_91ef6218b8df1f7033ab418c | window | T2 | no | no | yes | yes | yes | 0.333 | 0.776 | refused_or_blocked |
| gjv1_92e1672d5f59dfc70a0b5b7a | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_96630cf860d2fc33266a9e02 | window | T2 | no | no | no | yes | no | 0.333 | 0.582 | runaway |
| gjv1_97a6ac3e903516d8ddfa590d | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_9f86c043afb21f95f7e92bb3 | aggregate | T1 | yes | yes | yes | yes | yes | 0.667 | 0.790 | n/a |
| gjv1_9ff68e3578781ddaf670209f | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_a01cccb039241e4d6ea26d2c | aggregate | T1 | no | no | no | yes | no | 0.333 | 0.583 | runaway |
| gjv1_a1a97349d01dcb5170e3e125 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.993 | n/a |
| gjv1_a2cc8413c2ddedf87df006c3 | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_a32909984cc13c98e77ed944 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_a6372e0525903443ff68846b | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.926 | n/a |
| gjv1_a641a4acdc678d88b7c23e83 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_a7e59797b6d89ff12d9f5513 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_afbd28dcd992d26b85b20e91 | ranking | T3 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_b07ec17ef1b22f0130192f03 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_b63ed9d6b023b28c04bc0ab2 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_b901e2f0be48b7059a2928da | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_bd65154677b540ff64c17dfa | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_be6ee273f8c84d63e812b07f | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.994 | n/a |
| gjv1_c591e33247cce41440aaf3fd | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_c7d128262e4bac93df4716f9 | window | T2 | no | no | no | yes | no | 0.333 | 0.582 | runaway |
| gjv1_c83b925a4121e70b55da7ea9 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_c8d7ed90065c4876e49dc5cd | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.787 | n/a |
| gjv1_cb3afd9f8b971dd6d4e00efc | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.863 | n/a |
| gjv1_cf0a33332073925661609eb3 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_d16e895329a5597edc1c9490 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_d2652607e8a15e7d14af0e64 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_d3ff4fe19561e47c445f5682 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.994 | n/a |
| gjv1_d460002c26a423d1631e2482 | aggregate | T1 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_d5335166130128879dc1e4ed | window | T2 | no | no | yes | yes | no | 0.333 | 0.649 | runaway |
| gjv1_dbee4510ea1e0383de3f2c1d | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.787 | n/a |
| gjv1_de39d6c233250d2156baa5b2 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_e032bde975a5eb310356c719 | ranking | T3 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_e18526da4fa583f53ff1527b | aggregate | T1 | yes | yes | yes | yes | yes | 0.667 | 0.790 | n/a |
| gjv1_e427e329975e91c014234847 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.992 | n/a |
| gjv1_e4f8206cd9b0595011558816 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.789 | n/a |
| gjv1_e6696101c63650dffddfce21 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.855 | n/a |
| gjv1_e7a085ea2520dd171f391f01 | ranking | T3 | yes | yes | yes | yes | yes | 1.000 | 0.996 | n/a |
| gjv1_ea9c04051f82762b5e91d412 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_eea3b2b135648797d0f0a854 | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.862 | n/a |
| gjv1_f064fd6a857dda8cbf9bef88 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_f2dd8bac7c940fcb3b863684 | window | T2 | no | no | no | yes | no | 0.333 | 0.584 | refused_or_blocked |
| gjv1_f67d1609fac3a00b6cb4c996 | refusal | T4 | yes | n/a | n/a | yes | yes | 0.667 | 0.933 | n/a |
| gjv1_f8f20076dbbf2e19378ee684 | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_fc35b0ffe28cbb780f1cd817 | ranking | T3 | yes | yes | yes | yes | yes | 1.000 | 0.995 | n/a |
| gjv1_fe09d1c5686887bc50de635d | window | T2 | yes | yes | yes | yes | yes | 1.000 | 0.997 | n/a |
| gjv1_ffbb92a5a7460454e6eaf7fe | window | T2 | yes | yes | yes | yes | yes | 0.667 | 0.855 | n/a |
Invalid oracles
None.
This shareable report contains aggregate metrics, public task identifiers, fingerprints, and acceptance state. It excludes prompts, answers, database rows, executed queries, request headers, credentials, task slugs, raw oracle errors, and local episode paths.