This receipt lets anyone verify this run against the frozen suite — see Methodology.

suite-v2 | index-v4.2

Qwen3.6-27B

qwen3-6-27b__qwen3-6-27b-q4km-s2v5

Local Intelligence Index

LB-2026-07.2 | 25/22.5/22.5/22.5/7.5

43.2

±2.9 95% CI

Profile: Agentic / Knowledge / Instruction / Coding / Math

Agentic 8.3 / Knowledge 83.4 / Instruction 65.0 / Coding 25.5 / Math 26.6

Weighted headline profile: Agentic 25%, Knowledge 22.5%, Instruction 22.5%, Coding 22.5%, Math 7.5%.

Total run time

23.7 h

Data quality note: this run has 0 error(s) and 10 no-answer item(s).

Data warnings

tool_use chance_corrected differs from item-derived mean; stale/inconsistent run JSON
coding chance_corrected differs from item-derived mean; stale/inconsistent run JSON

Axis breakdown

Agenticworst axis

n=96 · errors=0 · no answer=0

8.3 ±7.4

Knowledge

n=400 · errors=0 · no answer=1

83.4 ±3.9

Instruction

n=294 · errors=0 · no answer=0

65.0 ±5.4

Coding

n=141 · errors=0 · no answer=0

25.5 ±7.1

Math

n=139 · errors=0 · no answer=9

26.6 ±7.2

Diagnostics · unweighted

Call formatting

n=330 · errors=0 · no answer=0

73.3 ±4.8

BFCL single-turn

— not measured

BFCL v3 multi-turn base — frozen snapshot

n=50 · errors=0 · no answer=0

22.0 ±12.0

BFCL multi-turn long-context

— not measured

RULER 32K

— not measured

Instruction-following decomposition

Strict = Termination × Conditional

Strict accuracy

65.0%

correct AND terminated / all

Termination rate

95.6%

terminated / all

Conditional accuracy

68.0%

correct / terminated

Outputs that hit the answer-token cap are counted incorrect; this prevents non-terminating generations from getting credit for matching required tokens inside a runaway response.

Manifest

model: qwen35
quant: Q4_K_M
runtime: llama.cpp b9852/fd1a05791 · KV k=f16,v=f16 · ctx 32,768
hardware: NVIDIA GeForce RTX 5090 (31.8 GB) · Windows-11-10.0.26200-SP0
os: Windows-11-10.0.26200-SP0
lane: bounded-final-v2
thinking_mode: n/a
caps: max_tokens_math: 0, max_tokens_mcq: 16384, thinking_budget: 8192
sampling: temp 0 | top_p n/a | top_k 1 | min_p n/a | seed 1234 | effort n/a | knowledge: max 16384 | instruction: max 16384 | coding: max 16384 | math: max 16384 | bfcl_multi_turn_base: max 16384 | tc_json_v1: max 16384
tokens: 6,525,242 prompt / 5,921,541 completion / 12,446,783 total
tokens-to-answer: 3,287 median / 9,380 p95
tok/s: 69.4
total run time: 23.7 h
est cost: n/a
n_items: 1,457
n_errors: 0
n_no_answer: 10

Serving performance

NVIDIA GeForce RTX 5090 (31.8 GB) · Windows-11-10.0.26200-SP0

prefill

3,112.1 tok/s

decode

74.4 tok/s

TTFT proxy

640.9 ms

prompt processing before first token — non-streaming harness, lower bound

coverage

93.4%

prompt median / p95

640.9 ms / 2,249.7 ms

predicted median / p95

43,212.2 ms / 130,836.7 ms

bench	prefill	decode	prompt median	n
amo	1,412 tok/s	76.4 tok/s	171.3 ms	39
bfcl_multi_turn_base	3,400.4 tok/s	74.3 tok/s	2,500.5 ms	50
bigcodebench_hard	3,345.6 tok/s	77.1 tok/s	1,306.4 ms	148
ifbench	3,163.1 tok/s	75.5 tok/s	935.5 ms	294
mmlu_pro	2,948.9 tok/s	72.8 tok/s	572.1 ms	400
olymmath_hard	874.5 tok/s	72.5 tok/s	169.2 ms	100
tc_json_v1	3,115.1 tok/s	77.8 tok/s	566.5 ms	330

Source: llama.cpp server timings.