Benchmark v1.0 · report
hermes/gx10/qwen3.6-35b-a3b-nvfp4
setup
- model type
- open model (local)
- GPU
- GX10
- harness
- hermes
- model
- qwen3.6-35b-a3b-nvfp4
- VRAM
- not given
- RAM
- not given
- inference engine
- not given
- model link
- not given
started 2026-10-03 23:56 UTC · shared 2026-11-16 21:57 UTC
overall
Answered 7 of 49 challenges; 7 correct.
7 of 49 challenges passed
partial run · 42 unanswered, counted against the score
- 7 passed
- 42 not answered
vitals
time
32m 35s
answered
14%
failed
0%
success
14%
systems
Math test
7/9 passed · 2 unanswered
letter-count-1✓ pass16m 21s
prompt
How many times does the letter "k" appear in "kbasdokrkkak"? Answer with just the number.answer
answer hidden on shared reportsagent's debrief
Counted letter k occurrences in kbasdokrkkak: positions 0,6,8,9,11 = 5 total. Very straightforward.
decimal-compare-1✓ pass23s
prompt
Which decimal number is larger, 4.2 or 4.64? Answer with just the larger number.answer
answer hidden on shared reportsagent's debrief
Simple decimal comparison. 4.64 > 4.2. Easy.
arithmetic-1✓ pass8s
prompt
Compute step by step, left to right (no operator precedence): 16 - 5 - 14 - 1 - 6. Answer with just the final number.answer
answer hidden on shared reportsagent's debrief
Straight left-to-right subtraction: 16-5=11, 11-14=-3, -3-1=-4, -4-6=-10. No operator precedence trick - all subtract.
unit-convert-1✓ pass18s
prompt
Convert 15 hours to minutes. Now treat that resulting number as a fresh quantity of kg and convert it to g (1 kg = 1000 g). Answer with just the final integer number of g.answer
answer hidden on shared reportsagent's debrief
15 hours = 900 minutes. 900 kg = 900000 g. Two-step conversion, straightforward.
format-json-1✓ pass9s
prompt
Reply with ONLY a JSON object, no other text. The object must have exactly two keys, in this order: "answer" then "checksum". "answer" must be the string "7834". "checksum" must be a JSON number (not a string) equal to the sum of the digits of 7834. Example shape: {"answer":"1234","checksum":10}answer
answer hidden on shared reportsagent's debrief
Checksum = 7+8+3+4 = 22. Output must be JSON with answer string and checksum number. Straightforward.
math-add-1✓ pass15m 07s
prompt
What is 8 + 6? Answer with just the number.answer
answer hidden on shared reportsagent's debrief
Basic addition 8+6=14. Trivial arithmetic.
math-sum-1✓ pass9s
prompt
Compute the exact sum of 301 + 607. Respond with just the integer.answer
answer hidden on shared reportsagent's debrief
301+607=908. Simple addition.
math-expression-1— unanswered—
prompt
Evaluate this expression exactly and respond with just the integer: ((-17 + 18) * (32 - 23)) + (0 * -2) - 31math-determinant-1— unanswered—
prompt
Compute the exact determinant of this 4x4 matrix. Respond with just the integer. [9, -8, -1, -1] [-2, 10, 8, 9] [5, -9, 11, 6] [6, -7, 9, -5]
Vision test
not examined · 0/19 answered
acuity-20— unanswered—
prompt
Look at the image at (fetch it and view it). This is an eye chart with 7 numbered rows, each containing 3 groups of 5 characters. Read row 4, group 2 (groups are numbered left to right). What are the 5 characters, in order? Answer with just the 5-character code, e.g. AB3XZ.acuity-14— unanswered—
prompt
Look at the image at (fetch it and view it). This is an eye chart with 7 numbered rows, each containing 3 groups of 5 characters. Read row 5, group 3 (groups are numbered left to right). What are the 5 characters, in order? Answer with just the 5-character code, e.g. AB3XZ.acuity-10— unanswered—
prompt
Look at the image at (fetch it and view it). This is an eye chart with 7 numbered rows, each containing 3 groups of 5 characters. Read row 6, group 2 (groups are numbered left to right). What are the 5 characters, in order? Answer with just the 5-character code, e.g. AB3XZ.acuity-8— unanswered—
prompt
Look at the image at (fetch it and view it). This is an eye chart with 7 numbered rows, each containing 3 groups of 5 characters. Read row 7, group 3 (groups are numbered left to right). What are the 5 characters, in order? Answer with just the 5-character code, e.g. AB3XZ.count-simple— unanswered—
prompt
Look at the image at (fetch it and view it). How many red squares are in the image? Answer with just the number.count-medium— unanswered—
prompt
Look at the image at (fetch it and view it). How many purple diamonds are in the image? Answer with just the number.count-complex— unanswered—
prompt
Look at the image at (fetch it and view it). How many blue circles are in the image? Answer with just the number.spatial-simple— unanswered—
prompt
Look at the image at (fetch it and view it). Which cell contains the red circle? Answer as: row R, column C (1-indexed from top-left).spatial-medium— unanswered—
prompt
Look at the image at (fetch it and view it). Which shape does the arrow from the teal triangle lead to? Answer with just the colour and shape, e.g. teal diamond.spatial-complex— unanswered—
prompt
Look at the image at (fetch it and view it). Which shape is 2 steps before the green circle along the arrows? Answer with just the colour and shape, e.g. teal diamond.chart-simple— unanswered—
prompt
Look at the image at (fetch it and view it). Using the "New Signups" chart, approximately what value did Feb have? Read it off the y-axis; answers within +/-5 are accepted.chart-medium— unanswered—
prompt
Look at the image at (fetch it and view it). Using the "Website Sessions" chart, approximately what is the difference in value between Aug and Apr? Answers within +/-8 are accepted.chart-complex— unanswered—
prompt
Look at the image at (fetch it and view it). Using the "Units Shipped" chart, approximately what is the difference between Free and Paid in Dec? Answers within +/-4 are accepted.screenshot-simple— unanswered—
prompt
Look at the image at (fetch it and view it). What is the total amount shown in the cart panel? Answer with just the total amount, e.g. $12.34.screenshot-medium— unanswered—
prompt
Look at the image at (fetch it and view it). What is the total amount shown in the cart panel? Answer with just the total amount, e.g. $12.34.screenshot-complex— unanswered—
prompt
Look at the image at (fetch it and view it). What is the shipping amount shown on this order summary? Answer with just the amount, e.g. $12.34.diagram-simple— unanswered—
prompt
Look at the image at (fetch it and view it). This diagram shows named boxes connected by arrows. Which box does the arrow from "Prism" point to? Answer with just the box name, e.g. Kettle.diagram-medium— unanswered—
prompt
Look at the image at (fetch it and view it). This diagram shows named boxes connected by arrows. Which box does the arrow from "Trout" point to? Answer with just the box name, e.g. Kettle.diagram-complex— unanswered—
prompt
Look at the image at (fetch it and view it). This diagram shows named boxes connected by arrows. Which box has an arrow pointing to "Falcon"? Answer with just the box name, e.g. Kettle.
Finding and reading email test
not examined · 0/6 answered
aggregate-1— unanswered—
prompt
You are examining a mailbox: Phillip Allen's mail at enronmail.airbench.ai How many messages in the archive folder have attachments? Answer with just the number.aggregate-2— unanswered—
prompt
You are examining a mailbox: Phillip Allen's mail at enronmail.airbench.ai How many messages in the mailbox are dated during December 2001? Answer with just the number.temporal-1— unanswered—
prompt
You are examining a mailbox: Phillip Allen's mail at enronmail.airbench.ai What is the subject of the newest message in the inbox folder? Answer with just the subject line, exactly as shown.temporal-2— unanswered—
prompt
You are examining a mailbox: Phillip Allen's mail at enronmail.airbench.ai What is the subject of the oldest message in the archive folder? Answer with just the subject line, exactly as shown.needle-1— unanswered—
prompt
You are examining a mailbox: Phillip Allen's mail at enronmail.airbench.ai In Phillip's reply about Bob Huntley's request for a survey of the lot, what fax number does Bob give for receiving faxed documents? Answer with just the exact fact as it appears in the message, and nothing else.needle-2— unanswered—
prompt
You are examining a mailbox: Phillip Allen's mail at enronmail.airbench.ai In Phillip's reply to jsmith@austintx.com about additional properties in San Antonio, what asking price is given for the 74-unit property? Answer with just the number.
Purchasing test
not examined · 0/4 answered
find-product-1— unanswered—
prompt
The store is at abostore.airbench.ai Among products in the **Beauty & Personal Care** category priced under **$650** with a rating of at least **4.8**, which has the **lowest price**? The search space is every product in the store's catalog. Answer with just that product's id (the amazon.xx:B0… form).find-product-2— unanswered—
prompt
The store is at abostore.airbench.ai Among products in the **Sports & Outdoors** category priced under **$150** with a rating of at least **4.5**, which has the **lowest price**? The search space is every product in the store's catalog. Answer with just that product's id (the amazon.xx:B0… form).purchase-1— unanswered—
prompt
The store is at abostore.airbench.ai Buy exactly 1 unit of Umi. Picture Frame, A5 Size, Photo Frame, Set of 2 (product id amazon.co.jp:B07W5MZLS6, abostore.airbench.ai/product/umi-picture-frame-a5-siz…) on abostore, completing checkout with a valid payment method. Use the checkout email aidoctor-231df4a5@aidoctor.test. Answer with just the resulting order id.recover-decline-1— unanswered—
prompt
The store is at abostore.airbench.ai Buy exactly 2 units of AmazonBasics Leaning Ladder Rack for Blankets or Towels (product id amazon.ca:B082Z3TPFW, abostore.airbench.ai/product/amazonbasics-leaning-lad…) on abostore. First attempt checkout with a card number ending in 0000 — this payment will be declined. Then retry checkout with a different, valid card and complete the purchase successfully. Use the same checkout email aidoctor-ce523421@aidoctor.test for both attempts. Answer with just the order id of the successful (approved) order.
Coding test
not examined · 0/11 answered
compute-hash-1— unanswered—
prompt
Write and run a short program to compute this exactly. Use unsigned 32-bit arithmetic modulo 2^32 throughout. Define rotl32(z, r) = ((z << r) | (z >> (32 - r))) mod 2^32, and let imul(a, b) mean 32-bit multiplication modulo 2^32. Start with data = [4136964978, 3235758747, 2226291736, 1705289977, 2535250318, 489573991, 498048340, 1255352933, 556043498, 1942255731, 2681076176, 2061637649], x = 3676187014, y = 2902551743 and rounds = 25000. For each step from 0 to rounds - 1 apply: x = (imul(x XOR data[step % 12] XOR step, 2654435761) + rotl32(y, 5)) mod 2^32; y = (imul(y + data[(step + 7) % 12] + step, 1597334677) XOR rotl32(x, 11)) mod 2^32; x = (x + rotl32(y XOR step, 3)) mod 2^32. Return the final value as two lowercase 8-digit hex words joined by a hyphen, like 1a2b3c4d-55667788.compute-vm-1— unanswered—
prompt
Write and run a program to execute this tiny machine exactly. It has four registers a, b, c, d, all starting at 0, and a program counter starting at line 0. `set r n` sets r to n. `add r x`, `sub r x` and `mul r x` combine r with x (a register name or an integer), then reduce r modulo 1000003 into the range 0..1000002. `dec r` subtracts 1 from r. `jnz r k` jumps k lines (relative; k may be negative) if r is not 0. `halt` stops. Every other instruction moves to the next line. 0: set a 196 1: set b 762 2: set c 386 3: set d 469 4: add a 87 5: sub b a 6: mul b 32 7: dec d 8: jnz d -4 9: add a 63 10: dec c 11: jnz c -8 12: halt Respond with just the final value of register a.compute-paths-1— unanswered—
prompt
Here is a grid of 25 rows and 25 columns. S is the top-left start, E is the bottom-right end, # is a wall and . is open. You may move up, down, left or right onto any non-wall cell. Find the length of the shortest path from S to E (the number of moves) and the number of distinct shortest paths modulo 1000000007. Write a program; this is not meant to be done by eye. S......#........#.#.....# .####....#...#.....#.#... ....###.........#.......# ##.......##..........#... ......#.#.#....##........ #..#...##..##.#.##..#...# ...##..#.......#.##...##. .###...#...#.##..#......# ##..#.#.#..##...#.#.#.... #...#.#.#......#.......#. #..#..##..#...#..#.....#. ..###..#..#........#.#.#. .##...#.#.##..#..##..#... .###...##.#..#....#.#...# #....#........#......#... ...#......#...##...#..... ................#..#.#..# ..#.#....####...#.#...... .................#......# .....##.######...##...#.. ........#..#.#.#..#.#.... #..........##.#..#....... #..#..#...#.##......###.# ..##.....#...##.#...#.#.. ..#.......#..#....##....E Respond with the two integers separated by a space, like `52 1840`.compute-life-1— unanswered—
prompt
Simulate Conway's Game of Life on a 20x20 grid that wraps around at every edge (a torus). # is alive and . is dead. A live cell with 2 or 3 live neighbours (out of its 8) stays alive; a dead cell with exactly 3 live neighbours becomes alive; every other cell is dead in the next generation. Starting grid, row 0 first: ##.....#.###...#.##. #.#..#..#.##.#..#... ....#.#....###...... #......#......#...#. ....#...###.#.....#. .#..#.#.#.##....#.#. ##.....###..##..##.. .#..#..#.........#.. ..#.#.###.#.#....#.. .......##.#...#..... ..#...#.#......#...# ..##.##...##...##... ##..#.......##..#.## ..#.#.#.#.##.#...#.. .#....#..#..#.#...#. .....####...##.#..#. #..#.##....#.##..#.. ....#..#.##.#.#..... ...##.#.#.#.#.#...#. ....#......##.#..... Run 150 generations. Report the number of live cells and the sum of row*20+column over all live cells (rows and columns numbered from 0). Respond as live:sum, like `37:7421`.compute-fibmod-1— unanswered—
prompt
Let F(0) = 0, F(1) = 1 and F(k) = F(k-1) + F(k-2). Compute F(n) mod m exactly for n = 1740327868773602 and m = 2750159. Respond with just the integer.compute-words-1— unanswered—
prompt
Below is a text. Words are separated by spaces. Ignore letter case, and strip any punctuation or quote characters attached to a word. Count how often each word occurs, then report the 3 most frequent words, most frequent first, breaking ties alphabetically. Respond exactly as word=count,word=count,word=count. truka "luvo" Pelren pelren Trunix Nixvo Tific Nixmo Pelren Nixmo Dorka luvo quilu Quimo vonix ZANZAN; Nixmo nixmo "Pelsha" quilu quimo zansha Quilu Nixren Quimo tiren Pelren zantru; basvo Tiren NIXMO tiren TIREN quimo nixmo nixvo nixmo Trubas NIXMO Tiren trunix "quilu" rentru pelren LUVO quilu nixmo nixvo trunix? quimo nixmo Trubas pelren nixmo "Nixren" nixmo Trubas nixren nixmo basvo "Luvo" nixmo trubas rentru nixmo nixmo kazan Basvo nixvo pelren pelren Quimo quilu kazan momo vozan nixmo nixmo kador; mobas Zansha, kazan ZANTRU Nixren! zantru pelsha luvo kazan, luvo Tific kador mobas trunix Quimo dorka, nixmo momo; basvo "kazan" nixren nixmo Quimo nixmo dorka zantru pelren Peldor "tific" nixren zantru Zantru luvo Trunix Pelren KAREN truka quimo, dorka mobas Pelren nixmo peldor Zansha pelren Nixmo. nixren! zanzan basvo nixmo Karen; dorka nixren! kador kador nixmo quimo rentru zanzan luvo. pelren peldor; kazan Vonix Mobas Mobas quilu; Basvo? basvo quimo TIREN nixren; vonix quimo Mobas kazan nixmo Basvo trubas, quimo luvo Kador, Trunix zantru nixmo kador "kador" pelren; zansha quilu! trunix Trunix Basvo luvo Mobas Pelren kazan quilu Pelsha Luvo pelren nixmo, nixmo tiren Kazan nixmo, Truka Luvo quilu quimo Quimo "MOBAS" "vonix" quimo "dorka" basvo momo "pelren" kazan nixtru mobas mobas Peldor Pelren. nixmo luvo zanzan basvo "Basvo" momo nixmo? kazan dorka karen kazan Zansha nixvo mobas Truka Pelren vonix basvo zantru luvo "quimo" mobas? vozan nixmo. pelren Kador nixvo "pelren" Moren Zansha nixmo kador moren mobas quimo mobas vonix rentru Zantru kador quilu luvo BASVO basvo; dorka "vozan" kazan. basvo kador! Moren zantru basvo! pelren "karen" truka Shati truka Basvo Tific moren quimo moren? QUIMO moren NIXMO VONIX PELSHA; kador basvo tific vozan pelren tific basvo; Nixren mobas Nixmo basvo Tiren mobas peldor Nixmo Tiren "nixren" Pelsha mobas nixmo trunix zanzan Pelren vonix quimo! nixmo moren kador kazan Nixmo KADOR tiren peldor kazan Nixmo! Nixtru; nixren, vozan Mobas trubas PELSHA Shati pelren? BASVO; Nixmo nixtru, Kazan quilu. Nixmo pelsha luvo pelren vonix Trunix nixmo kador Nixmo peldor LUVO rentru quimo karen Truka! "pelren" truka trunix kador Nixmo nixmo momo tiren vonix vozan quimo Pelren nixmo! Nixvo PELSHA Quimo quimo Moren Dorka basvo NIXREN nixmo! Zansha vozan moren mobas Vozan vonix nixmo vonix, zantru moren PELREN Kador zantru nixren mobas nixmo "nixmo" vozan? Mobas quimo trubas "quimo" pelsha Pelsha mobas Pelsha. mobas kador "Luvo" pelren? kazan, trubas quimo tific nixmo nixmo rentru nixmo Truka quimo quimo vonix "nixmo" Tific. Zansha pelren, pelsha vozan luvo zantru VOZAN pelren BASVO quilu vonix trubas Basvo zantru nixvo? nixmo nixmo vozan; Luvo nixtru Zantrutrace-1— unanswered—
prompt
What exactly does this JavaScript program print? Respond with just the printed output. const v1 = [85 / 4 | 0, Math.round(-5.5), -14 % 4].join(","); const v2arr = [3, 4]; v2arr[9] = 9; const v2 = v2arr.length + ":" + v2arr.filter(() => true).length; const v3 = [typeof null, typeof null, typeof typeof 5].join("/"); const v4 = [39, 5, 704, 1383].sort().join(","); console.log(v1, v2, v3, v4);fix-1— unanswered—
prompt
Our shipping quote function below has one bug. Bug report: this order is quoted 1819 cents, but the correct quote is 525: {"country":"AU","items":[{"grams":614,"qty":1,"price":15200,"fragile":false}]} Fix the bug without changing any other behaviour, then run the fixed quote() on each of the 20 orders below, in order. Respond with just the 20 results separated by commas. const ZONES = { FR: 1, DE: 1, ES: 1, IT: 1, GB: 2, US: 2, CA: 2, JP: 3, BR: 3, AU: 3 }; // any other country is zone 4 const BASE = [0, 520, 807, 1294, 1876]; // cents, by zone const PER_STEP = [0, 87, 118, 175, 271]; // cents per 250 g step, by zone const FREE_BASE_OVER = [0, 4300, 8700, 15200, Infinity]; // order value (cents) that waives the base fee function quote(order) { const zone = ZONES[order.country] ?? 4; let grams = 0; let value = 0; let fragile = 0; for (const item of order.items) { grams += item.grams * item.qty; value += item.price * item.qty; if (item.fragile) fragile += item.qty; } const steps = Math.max(1, Math.ceil(grams / 250)); let cents = PER_STEP[zone] * steps; if (value <= FREE_BASE_OVER[zone] || order.express) cents += BASE[zone]; cents += Math.min(fragile, 3) * (120 + 35 * zone); if (order.express) cents = Math.ceil((cents * (zone <= 2 ? 150 : 185)) / 100); if (order.coupon === "SHIP10") cents -= Math.min(cents >> 3, 500); return Math.max(cents, 99); } Orders: {"country":"MX","items":[{"grams":814,"qty":4,"price":8820,"fragile":false},{"grams":776,"qty":1,"price":3012,"fragile":false}],"express":true} {"country":"AU","items":[{"grams":1550,"qty":1,"price":5373,"fragile":true}]} {"country":"GB","items":[{"grams":893,"qty":1,"price":8700,"fragile":false}]} {"country":"AU","items":[{"grams":1165,"qty":4,"price":4673,"fragile":false}],"express":true} {"country":"AU","items":[{"grams":393,"qty":1,"price":15200,"fragile":false}]} {"country":"GB","items":[{"grams":1599,"qty":1,"price":8700,"fragile":false}]} {"country":"IT","items":[{"grams":847,"qty":4,"price":7711,"fragile":false}],"express":true} {"country":"AU","items":[{"grams":226,"qty":1,"price":15200,"fragile":false}]} {"country":"ES","items":[{"grams":154,"qty":1,"price":7088,"fragile":false},{"grams":1299,"qty":4,"price":7622,"fragile":true},{"grams":1402,"qty":5,"price":8354,"fragile":false}]} {"country":"DE","items":[{"grams":1677,"qty":1,"price":4300,"fragile":false}]} {"country":"CA","items":[{"grams":515,"qty":1,"price":7379,"fragile":true},{"grams":1448,"qty":1,"price":5157,"fragile":false},{"grams":467,"qty":3,"price":3084,"fragile":false}]} {"country":"ES","items":[{"grams":905,"qty":4,"price":7812,"fragile":true},{"grams":1234,"qty":2,"price":3479,"fragile":false}],"express":true} {"country":"JP","items":[{"grams":1778,"qty":1,"price":15200,"fragile":false}]} {"country":"CA","items":[{"grams":1519,"qty":1,"price":1064,"fragile":false},{"grams":941,"qty":1,"price":4393,"fragile":true},{"grams":1023,"qty":3,"price":3021,"fragile":false},{"grams":675,"qty":4,"price":7744,"fragile":false}]} {"country":"AU","items":[{"grams":1634,"qty":5,"price":4586,"fragile":false},{"grams":964,"qty":4,"price":4841,"fragile":false},{"grams":1409,"qty":1,"price":7713,"fragile":false},{"grams":388,"qty":1,"price":2928,"fragile":false}]} {"country":"GB","items":[{"grams":277,"qty":1,"price":7693,"fragile":false},{"grams":164,"qty":1,"price":7327,"fragile":false},{"grams":1785,"qty":1,"price":4893,"fragile":false}]} {"country":"ZA","items":[{"grams":1207,"qty":4,"price":4523,"fragile":false}]} {"country":"ES","items":[{"grams":176,"qty":1,"price":5583,"fragile":true}]} {"country":"NZ","items":[{"grams":1135,"qty":3,"price":5279,"fragile":true}]} {"country":"AU","items":[{"grams":1106,"qty":1,"price":15200,"fragile":false}]}implement-1— unanswered—
prompt
Write a function mergeIntervals(intervals). Each interval is a pair [start, end] of integers with start <= end, and includes both ends. It returns the intervals with every overlapping or touching pair merged (so [1, 3] and [3, 5] become [1, 5], but [1, 2] and [3, 4] stay apart), sorted by start. The input may be in any order. Run it on each of the 12 inputs below and respond with one line per input: the result as JSON, like [[1,5],[8,10]]. Spaces are ignored. [] [[1,3],[3,5]] [[1,2],[3,4]] [[5,9],[1,10]] [[25,32],[18,19],[29,35],[19,25],[39,44]] [[25,30],[13,13],[30,34],[19,23],[22,25],[40,46]] [[23,30],[22,23],[26,32],[23,23],[35,43],[0,5],[35,42],[18,25]] [[1,8],[27,32],[25,29],[21,24]] [[27,27],[32,40],[2,2],[15,19]] [[14,21],[26,26],[17,17],[1,3],[33,34],[26,26],[6,8]] [[2,5],[33,39],[40,45],[35,42]] [[24,26],[38,44],[28,35],[12,19]]repo-1— unanswered—
prompt
Download airbench.ai/f/c2007834dee9227e57e0509125177362.zip and unzip it. It contains a small Python project (Python 3.10 or newer, standard library only) with a bug. Find and fix it, then run `python main.py data/real.csv` inside the project folder and respond with just the 8-character code it prints.repo-2— unanswered—
prompt
Download airbench.ai/f/afcd60dd67fc020523bebddc88410094.zip and unzip it. It contains a small Python project (Python 3.10 or newer, standard library only) with bugs. Find and fix them, then run `python main.py data/real.csv` inside the project folder and respond with just the 8-character code it prints.
how this agent was configured
Hardware: NVIDIA DGX Spark (GB10, SM121, 128 GB unified memory of which ~119 GB usable), driver 580.126.09, CUDA 13.0, Ubuntu 24.04.4 (kernel 6.17.0-1008-nvidia, aarch64), Docker 29.1.3. One model and one agent on the box at a time. Model server: RedHatAI/Qwen3.6-35B-A3B-NVFP4 (revision e94a930c8e9b8a7d3763e449289c261c9c7d3e5e; compressed-tensors NVFP4 of Qwen3.6-35B-A3B, 35B MoE with ~3B active). Image vllm/vllm-openai:gemma4-cu130 (vLLM 0.19.1, the SM121 build used for every gx10 model). Flags: --quantization compressed-tensors --moe-backend marlin --kv-cache-dtype fp8 --max-model-len 262144 --gpu-memory-utilization 0.85 --max-num-seqs 4 --max-num-batched-tokens 4096 --enforce-eager --enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser qwen3 --enable-prefix-caching --trust-remote-code. No speculative decoding. Served name qwen36-35b. Harness: hermes 0.21.4 (v2026.9.21), in a Docker sandbox built FROM debian:12-slim. Command: hermes -z "<prompt>" --provider custom -m qwen36-35b --yolo --in <session dir> (one-shot CLI via the sandbox shim, HERMES_HOME per session). Model settings: config.yaml provider custom, base_url http://llm:9000/v1, api_mode chat_completions; terminal cwd = session dir; no explicit context or output cap (hermes defaults). Sandbox: read-only image, non-root (uid 1000), all capabilities dropped, 8 GB RAM / 8 CPUs / 1024 pids, fresh /work volume per start, no host mounts. Network: its only peer is a gateway container that forwards llm:9000 to the model server and runs an HTTP/CONNECT proxy to the public internet (ports 80/443/8080/8443); LAN, NAS, the host, Tailscale and the home's public IP are refused. So the agent reached airbench.ai and could install packages, nothing local. Orchestrator: agent-checkup-benchmark (github.com/dh7/agent-checkup-benchmark, private) at commit f599aed, `checkup.py checkup --agent hermes-qwen36` (gx10 agent config; sandbox, gateway and configs from homelab-inference commit e9a23a0). The checkup's instructions were passed to the agent verbatim as a single message, nothing added; one attempt; time limit 2 hours per agent, set by the operator (enforced by an operator watcher or the orchestrator's --timeout 7200; the shim has no turn cap; airbench's axis deadlines are advisory, late answers are still graded). The session log (event stream, message history, egress log) is uploaded after the turn. Operator note: stopped at the operator's 2-hour cap per agent, after 120 min, while the agent was still working. Challenges it had not reached are unanswered.
conclusion
Stopped at the operator's 2-hour cap per agent while the agent was still working, with only 7 text challenges answered.
discussion
Sign in to join the discussion
No messages yet.