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

suite-v2 | index-v4.2

Qwopus3.6 27B v2 MTP

qwopus3-6-27b-v2-mtp__qwopus3-6-27b-v2-mtp-q4km-s2v5

Local Intelligence Index

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

42.1

±3.0 95% CI

Profile: Agentic / Knowledge / Instruction / Coding / Math

Agentic 9.4 / Knowledge 80.4 / Instruction 57.8 / Coding 29.8 / Math 25.9

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

Total run time

23.6 h

Data quality note: this run has 0 error(s) and 49 no-answer item(s).

Data warnings

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

Axis breakdown

Agenticworst axis

n=96 · errors=0 · no answer=0

9.4 ±8.0

Knowledge

n=400 · errors=0 · no answer=12

80.4 ±4.2

Instruction

n=294 · errors=0 · no answer=0

57.8 ±5.8

Coding

n=141 · errors=0 · no answer=0

29.8 ±7.8

Math

n=139 · errors=0 · no answer=37

25.9 ±7.2

Diagnostics · unweighted

Call formatting

n=330 · errors=0 · no answer=1

73.9 ±4.8

BFCL single-turn

— not measured

BFCL v3 multi-turn base — frozen snapshot

n=50 · errors=0 · no answer=0

24.0 ±12.0

BFCL multi-turn long-context

— not measured

RULER 32K

— not measured

Instruction-following decomposition

Strict = Termination × Conditional

Strict accuracy

57.8%

correct AND terminated / all

Termination rate

93.9%

terminated / all

Conditional accuracy

61.6%

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: 5,690,781 prompt / 5,429,223 completion / 11,120,004 total
tokens-to-answer: 1,623 median / 16,384 p95
tok/s: 63.8
total run time: 23.6 h
est cost: n/a
n_items: 1,457
n_errors: 0
n_no_answer: 49

Serving performance

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

prefill

2,729.1 tok/s

decode

71 tok/s

TTFT proxy

396.1 ms

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

coverage

93.4%

prompt median / p95

396.1 ms / 1,874.1 ms

predicted median / p95

22,347.4 ms / 216,472.4 ms

bench	prefill	decode	prompt median	n
amo	1,250.6 tok/s	66.8 tok/s	197.1 ms	39
bfcl_multi_turn_base	3,324.4 tok/s	72.5 tok/s	1,976.4 ms	50
bigcodebench_hard	2,711.9 tok/s	66.6 tok/s	209.7 ms	148
ifbench	2,881.6 tok/s	76.4 tok/s	381.7 ms	294
mmlu_pro	2,582.8 tok/s	69.5 tok/s	362.4 ms	400
olymmath_hard	945.7 tok/s	71.6 tok/s	161.4 ms	100
tc_json_v1	2,657.5 tok/s	68.7 tok/s	471.4 ms	330

Source: llama.cpp server timings.