AgentBench / results
← All models

Qwen3.8 27B GSQ-RCO

3 tested weight formats · Explore results, then inspect the answers.

Choose a tested configuration

Get this tested model

IQ2_S · 9.26 GB download

Links follow the selected quant and configuration.

Setup, memory & provenance · IQ2_S

IQ2_S · 9.26 GB of weights · rocm · context 16384 · GPU layers 99

Large · 20B and above · dense · 26.9B total

Memory fit has not been verified for this configuration.

completed

Recorded settings and thinking verification

Temperature 0 · seed 42 · output limit 4096 · reasoning effort unrecorded

RAM labels describe the test computer, not this model’s memory requirement. Speed comparisons keep the 32 GB and 128 GB runs separate.

Runtime build: unrecorded · Local · test machine with 128 GB RAM · concurrency: 1

Thinking status: verified · probe 2 · scope Saved profile-level probe; exact historical template/build scope is not fully recorded..

Requested off does not prove an absence of reasoning. Saved checks do not establish unrecorded historical template or build details.

Overall across tasks

69.4 / 100

Nine categories, equally weighted · 9/9 eligible

Complete coverage for this setup. Quality only; speed is separate. How the index works →

What does it do well?

Select a category to explore the answers. Bars show what happened on every planned task. Incorrect answers and failures to finish both earn zero; technical problems and untested tasks stay unscored.

Exact scores and task times by suite

Each suite has its own metric and fixed task bank. The balanced index weights categories equally; individual scores and evidence remain available below.

Test suiteThinking onDetails
Everyday coding tasksTask pass rate
Pass rate
13/16 · 81%
Answers completed
13/16 · 81%
Correct / completed
13/13 · 100%
13 full credit3 failed to finish
16/16 scored30.1s median task time · 16 measured
Larger practical tasksMean task credit
Average credit
90.9%
Full-credit rate
10/11 · 91%
Answers completed
11/11 · 100%
Full credit / completed
10/11 · 91%
10 full credit1 incorrect
11/11 scored164.6s median task time · 11 measured
Logic and number puzzlesMean puzzle credit
Average credit
63.5%
Full-credit rate
21/34 · 62%
Answers completed
29/34 · 85%
Full credit / completed
21/29 · 72%
21 full credit2 partial credit6 incorrect5 failed to finish
34/34 scored125.2s median task time · 34 measured
Published hard problemsMean answer credit
Average credit
59.7%
Full-credit rate
6/11 · 55%
Answers completed
11/11 · 100%
Full credit / completed
6/11 · 55%
6 full credit2 partial credit3 incorrect
11/11 scored124.4s median task time · 11 measured
Following exact instructions · original campaignStrict prompt complianceNot planned
Reading without guessingExact-match accuracy
Pass rate
12/16 · 75%
Answers completed
14/16 · 88%
Correct / completed
12/14 · 86%
12 full credit2 incorrect2 failed to finish
16/16 scored18.2s median task time · 16 measured
Calculating from a passageNumeric exact-match accuracy
Pass rate
5/7 · 71%
Answers completed
5/7 · 71%
Correct / completed
5/5 · 100%
5 full credit2 failed to finish
7/7 scored15.6s median task time · 7 measured
Reading research papersAnswer token F1
Average credit
30.0%
Full-credit rate
0/8 · 0%
Answers completed
6/8 · 75%
Full credit / completed
0/6 · 0%
6 partial credit2 failed to finish
8/8 scored47.9s median task time · 8 measured
Resisting incorrect correctionsAll-branches pass rate
Pass rate
10/12 · 83%
Answers completed
12/12 · 100%
Correct / completed
10/12 · 83%
10 full credit2 incorrect
12/12 scored32.7s median task time · 12 measured
Following exact instructions · 128 GB campaignStrict prompt compliance
Pass rate
16/23 · 70%
Answers completed
17/23 · 74%
Correct / completed
16/17 · 94%
16 full credit1 incorrect6 failed to finish
23/23 scored62.4s median task time · 23 measured

Model analysis

This analysis covers this configuration across its recorded thinking conditions and test suites.

Use this tested GSQ-RCO setup with thinking for small-project coding and larger practical jobs: it succeeded on most of both on these tests. Research-paper questions remain a poor fit, and difficult reasoning needs checking. On this test machine, typical small coding tasks took 30.1 seconds, versus 164.6 seconds for larger work. Main-run and repeat coding results disagreed sharply, so check the setup on your own workload.

Evidence for this judgment
Everyday coding tasks · Thinking on
Tasks passed13 of 16 (81.3%)
Test coverage16 of 16 tasks scored
Tasks timed16 tasks
Typical time per task (median)30.1 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Larger practical tasks · Thinking on
Average score90.9/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes10 full credit · 0 partial · 1 no credit
Tasks timed11 tasks
Typical time per task (median)164.6 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Reading research papers · Thinking on
Average score30.0/100 · Answer token F1
Test coverage8 of 8 tasks scored
Task outcomes0 full credit · 6 partial · 2 no credit
Explore these test results →

View this model’s answers, code and tool activity →

Published hard problems · Thinking on
Average score59.7/100 · Mean answer credit
Test coverage11 of 11 tasks scored
Task outcomes6 full credit · 2 partial · 3 no credit
Explore these test results →

View this model’s answers, code and tool activity →

Logic and number puzzles · Thinking on
Average score63.5/100 · Mean puzzle credit
Test coverage34 of 34 tasks scored
Task outcomes21 full credit · 2 partial · 11 no credit
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking off versus on
Thinking off → on0.0% · Task pass rate → 81.3% · Task pass rate
Paired test coverage16 of 16 tasks compared in both modes
Tasks timed in both modes16 pairs
Average time with thinking on291.5 seconds
Average time with thinking off755.3 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking off
Tasks passed0 of 16 (0.0%)
Test coverage16 of 16 tasks scored
Tasks timed16 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Larger practical tasks · Thinking off versus on
Thinking off → on63.6/100 · Mean task credit → 90.9/100 · Mean task credit
Paired test coverage11 of 11 tasks compared in both modes
Tasks timed in both modes11 pairs
Average time with thinking on235.2 seconds
Average time with thinking off182.5 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Larger practical tasks · Thinking off
Average score63.6/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes7 full credit · 0 partial · 4 no credit
Tasks timed11 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Resisting incorrect corrections · Thinking off versus on
Thinking off → on8.3% · All-branches pass rate → 83.3% · All-branches pass rate
Paired test coverage12 of 12 tasks compared in both modes
Tasks timed in both modes12 pairs
Explore these test results →

View this model’s answers, code and tool activity →

Resisting incorrect corrections · Thinking on
Tasks passed10 of 12 (83.3%)
Test coverage12 of 12 tasks scored
Explore these test results →

View this model’s answers, code and tool activity →

Resisting incorrect corrections · Thinking off
Tasks passed1 of 12 (8.3%)
Test coverage12 of 12 tasks scored
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking off · sampled repetitions
Repetitions scored48 repetitions
Repetitions planned48 repetitions
Repetitions earning full credit48 repetitions
Questions with all repetitions scored16 questions
Distinct questions planned16 questions
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking on · sampled repetitions
Repetitions scored48 repetitions
Repetitions planned48 repetitions
Repetitions earning full credit48 repetitions
Questions with all repetitions scored16 questions
Distinct questions planned16 questions
Explore these test results →

View this model’s answers, code and tool activity →

Summary generated by GPT Astra · · Edited by Codex; agent reviewed by Codex root · publication authorized by user

About this summary
Analyst provider
openai
Generation route
Codex CLI · codex-cli 0.163.0 · ChatGPT account
Requested model
gpt-6-astra
Returned model
Not supplied; identity is recorded from the request only.
Exact provider snapshot
Exact provider snapshot not supplied.
Reasoning effort
high
Prompt version
model-summary-reader-v8
Generation time
2026-10-28T01:26:48.945091Z
Review time
2026-10-28T01:29:11.518570+00:00
Evidence reviewer
Codex root (agent)
Publication authorization
User · 2026-10-22T05:55:43.625712+00:00 · Go ahead and complete the remaining work. Spin up agents: luna, terra, sol, astra, as needed to get it done. Narrative summaries: these are the astra generated summaries? Go ahead and get those completed as well. Once all of this is done, you may push the completed result.
Originating evidence edition
dataset-161a0c0b3366764cd46aca8c85ef893157074bade44c7f42829d071766da607e
Analysis input hash
9c5457092adcfe185cc3bed1bf0343d06a8d68658becf1646658cc4d3dcc7da2
Summary revision
summary-10f18971357e507b6099b8cb26bc2b5896dabc4dfc9b3ad22f1e531db7d9f360

Generation and editorial authorship remain attached to this revision. Changing the default generator does not change this attribution.

Good at

Handles practical project work well with thinking enabled, succeeding on most small coding tasks and nearly all larger practical tasks. Its usefulness extends beyond isolated code snippets, although some small tasks did not finish.

Evidence for this judgment
Larger practical tasks · Thinking on
Average score90.9/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes10 full credit · 0 partial · 1 no credit
Incorrect completed answers1 tasks
Failed to finish0 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking on
Tasks passed13 of 16 (81.3%)
Test coverage16 of 16 tasks scored
Failed to finish3 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Usually keeps source-backed answers intact under incorrect challenges while accepting genuine source updates with thinking enabled. That makes it useful for conversations where evidence changes, rather than merely agreeing with the latest assertion.

Evidence for this judgment
Resisting incorrect corrections · Thinking on
Tasks passed10 of 12 (83.3%)
Test coverage12 of 12 tasks scored
Incorrect completed answers2 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Struggles with

Research-paper answers are a substantial weakness. With thinking, answer matching remained poor: completed responses were only partial matches, and some attempts did not finish. Without thinking, all answers in this test were incorrect.

Evidence for this judgment
Reading research papers · Thinking on
Average score30.0/100 · Answer token F1
Test coverage8 of 8 tasks scored
Task outcomes0 full credit · 6 partial · 2 no credit
Partially correct answers6 tasks
Failed to finish2 tasks
Full-credit answers0 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Reading research papers · Thinking off
Average score0.0/100 · Answer token F1
Test coverage8 of 8 tasks scored
Task outcomes0 full credit · 0 partial · 8 no credit
Incorrect completed answers8 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Difficult reasoning still needs checking: hard math/reasoning questions and generated puzzles had mixed outcomes with thinking. Some puzzle attempts failed to finish; others produced incorrect or only partly correct answers.

Evidence for this judgment
Published hard problems · Thinking on
Average score59.7/100 · Mean answer credit
Test coverage11 of 11 tasks scored
Task outcomes6 full credit · 2 partial · 3 no credit
Full-credit answers6 tasks
Incorrect completed answers3 tasks
Partially correct answers2 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Logic and number puzzles · Thinking on
Average score63.5/100 · Mean puzzle credit
Test coverage34 of 34 tasks scored
Task outcomes21 full credit · 2 partial · 11 no credit
Full-credit answers21 tasks
Incorrect completed answers6 tasks
Partially correct answers2 tasks
Failed to finish5 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Should you enable thinking?

Prefer the tested thinking-on setup for practical tasks. Main-run coding failed every task without thinking, yet coding repeats succeeded in both modes—a striking inconsistency. In the main run, thinking improved coding and larger-task success; coding was faster on average and larger jobs slower.

Evidence for this judgment
Everyday coding tasks · Thinking off versus on
Thinking off → on0.0% · Task pass rate → 81.3% · Task pass rate
Paired test coverage16 of 16 tasks compared in both modes
Tasks timed in both modes16 pairs
Average time with thinking on291.5 seconds
Average time with thinking off755.3 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking on
Tasks passed13 of 16 (81.3%)
Test coverage16 of 16 tasks scored
Tasks timed16 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking off
Tasks passed0 of 16 (0.0%)
Test coverage16 of 16 tasks scored
Tasks timed16 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Larger practical tasks · Thinking off versus on
Thinking off → on63.6/100 · Mean task credit → 90.9/100 · Mean task credit
Paired test coverage11 of 11 tasks compared in both modes
Tasks timed in both modes11 pairs
Average time with thinking on235.2 seconds
Average time with thinking off182.5 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Larger practical tasks · Thinking on
Average score90.9/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes10 full credit · 0 partial · 1 no credit
Tasks timed11 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Larger practical tasks · Thinking off
Average score63.6/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes7 full credit · 0 partial · 4 no credit
Tasks timed11 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Resisting incorrect corrections · Thinking off versus on
Thinking off → on8.3% · All-branches pass rate → 83.3% · All-branches pass rate
Paired test coverage12 of 12 tasks compared in both modes
Tasks timed in both modes12 pairs
Explore these test results →

View this model’s answers, code and tool activity →

Resisting incorrect corrections · Thinking on
Tasks passed10 of 12 (83.3%)
Test coverage12 of 12 tasks scored
Explore these test results →

View this model’s answers, code and tool activity →

Resisting incorrect corrections · Thinking off
Tasks passed1 of 12 (8.3%)
Test coverage12 of 12 tasks scored
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking off · sampled repetitions
Repetitions scored48 repetitions
Repetitions planned48 repetitions
Repetitions earning full credit48 repetitions
Questions with all repetitions scored16 questions
Distinct questions planned16 questions
Explore these test results →

View this model’s answers, code and tool activity →

Everyday coding tasks · Thinking on · sampled repetitions
Repetitions scored48 repetitions
Repetitions planned48 repetitions
Repetitions earning full credit48 repetitions
Questions with all repetitions scored16 questions
Distinct questions planned16 questions
Explore these test results →

View this model’s answers, code and tool activity →

Turn thinking off for straightforward passage questions when responsiveness matters: it finished sooner and was narrowly more accurate in these runs. That advice concerns finding answers in passages, not researching papers.

Evidence for this judgment
Reading without guessing · Thinking off versus on
Thinking off → on81.3% · Exact-match accuracy → 75.0% · Exact-match accuracy
Paired test coverage16 of 16 tasks compared in both modes
Tasks timed in both modes16 pairs
Average time with thinking on78.0 seconds
Average time with thinking off62.7 seconds
Median time with thinking on18.2 seconds
Median time with thinking off2.7 seconds
Explore these test results →

View this model’s answers, code and tool activity →

Reading without guessing · Thinking on
Tasks passed12 of 16 (75.0%)
Test coverage16 of 16 tasks scored
Tasks timed16 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Reading without guessing · Thinking off
Tasks passed13 of 16 (81.3%)
Test coverage16 of 16 tasks scored
Tasks timed16 tasks
Explore these test results →

View this model’s answers, code and tool activity →

Worth knowing

These GSQ-RCO artifacts are distinct from the original weights; differences cannot be attributed to quantization alone.

Evidence for this judgment

Source limitation
These are Qwen3.8 27B GSQ-RCO artifacts. They are kept separate from the original Qwen3.8 27B weights; differences cannot be attributed to quantization alone.

Publication redactions limit readers’ inspection of some evidence.

Evidence for this judgment

Source limitation
Some selected evidence has recorded publication redactions.

How to interpret these assessments and their limitations →

How do the weight formats compare?

Everyday coding tasks · Task pass rate · requested / displayed reasoning on. Size is the weight download, not RAM. Different runtimes and placement can also affect results.

IQ2_S · 9.26 GBLocal · test machine with 128 GB RAM · rocm · 99 GPU layers · Fully scoredThinking on13/16 passed16/16 scored
IQ2_S · MTP weights; drafting disabled · 9.61 GBLocal · test machine with 128 GB RAM · rocm · 99 GPU layers · Fully scoredThinking on16/16 passed16/16 scored
IQ3_XXS · MTP weights; drafting disabled · 10.44 GBLocal · test machine with 128 GB RAM · rocm · 99 GPU layers · Fully scoredThinking on16/16 passed16/16 scored

Qualified rows show the reason they are outside the rankings; they cannot establish superiority over a fully tested format. Open the setup to inspect context, runtime and controls.

Choosing a quant

Shared assessment of IQ2_S · MTP weights; drafting disabled, IQ2_S, IQ3_XXS · MTP weights; drafting disabled · 2026-10-28 UTC. The current results table can include newer formats.

In the main small-coding tests, both MTP-file setups passed every task in each thinking mode. The IQ2_S MTP file offers the smaller download of those two. The non-MTP IQ2_S setup scored lower in this main run, especially with thinking off; treat that as a result to investigate, not a settled judgment of its coding ability. These are quality-only comparisons of recorded setups, not evidence that MTP or quantization caused the differences. MTP drafting was disabled, and execution settings changed during recovery of the MTP-file runs.

Evidence for this judgment

Everyday coding tasks · Thinking requested on

Matching task controls and 16,384 context. These quality results describe the recorded setups; they do not isolate the effect of bit depth. Task times are not comparable for this group.

Weight formatTask pass rateTask timeDownloadInspect
IQ2_S · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared0/16 timed9.61 GBWeights, not required RAMAnswers and code
IQ2_SSelected formatPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
13/16 passed16/16 scoredNot compared16/16 timed9.26 GBWeights, not required RAMAnswers and code
IQ3_XXS · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared0/16 timed10.44 GBWeights, not required RAMAnswers and code

Everyday coding tasks · Thinking requested off

Matching task controls and 16,384 context. These quality results describe the recorded setups; they do not isolate the effect of bit depth. Task times are not comparable for this group.

Weight formatTask pass rateTask timeDownloadInspect
IQ2_S · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared0/16 timed9.61 GBWeights, not required RAMAnswers and code
IQ2_SSelected formatPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
0/16 passed16/16 scoredNot compared16/16 timed9.26 GBWeights, not required RAMAnswers and code
IQ3_XXS · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared0/16 timed10.44 GBWeights, not required RAMAnswers and code

For larger tool-driven work with reasoning requested on, consider IQ2_S: it earned the highest task credit while offering the smallest download. Both IQ2_S · MTP weights; drafting disabled and IQ3_XXS · MTP weights; drafting disabled scored lower.

Evidence for this judgment

Larger practical tasks · Thinking requested on

Matching task controls and 16,384 context. These quality results describe the recorded setups; they do not isolate the effect of bit depth. Task times are not comparable for this group.

Weight formatMean task creditTask timeDownloadInspect
IQ2_S · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
72.7/10011/11 scoredNot compared0/11 timed9.61 GBWeights, not required RAMAnswers and code
IQ2_SSelected formatPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
90.9/10011/11 scoredNot compared11/11 timed9.26 GBWeights, not required RAMAnswers and code
IQ3_XXS · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
72.7/10011/11 scoredNot compared0/11 timed10.44 GBWeights, not required RAMAnswers and code

With reasoning requested off, larger tool-driven tasks favored IQ3_XXS · MTP weights; drafting disabled. Its lead over IQ2_S · MTP weights; drafting disabled was narrow, making the latter's smaller download worth considering. Both outscored IQ2_S.

Evidence for this judgment

Larger practical tasks · Thinking requested off

Matching task controls and 16,384 context. These quality results describe the recorded setups; they do not isolate the effect of bit depth. Task times are not comparable for this group.

Weight formatMean task creditTask timeDownloadInspect
IQ2_S · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
81.8/10011/11 scoredNot compared0/11 timed9.61 GBWeights, not required RAMAnswers and code
IQ2_SSelected formatPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
63.6/10011/11 scoredNot compared11/11 timed9.26 GBWeights, not required RAMAnswers and code
IQ3_XXS · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
90.9/10011/11 scoredNot compared0/11 timed10.44 GBWeights, not required RAMAnswers and code

For strict wording and formatting requirements, IQ2_S · MTP weights; drafting disabled led with reasoning requested on; IQ3_XXS · MTP weights; drafting disabled led with reasoning requested off. The larger download is not a universal quality upgrade.

Evidence for this judgment

Following exact instructions · Thinking requested on

Matching task controls and 16,384 context. These quality results describe the recorded setups; they do not isolate the effect of bit depth. Task times are not comparable for this group.

Weight formatStrict prompt complianceTask timeDownloadInspect
IQ2_S · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
23/23 passed23/23 scoredNot compared0/23 timed9.61 GBWeights, not required RAMAnswers and code
IQ2_SSelected formatPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/23 passed23/23 scoredNot compared23/23 timed9.26 GBWeights, not required RAMAnswers and code
IQ3_XXS · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
20/23 passed23/23 scoredNot compared0/23 timed10.44 GBWeights, not required RAMAnswers and code

Following exact instructions · Thinking requested off

Matching task controls and 16,384 context. These quality results describe the recorded setups; they do not isolate the effect of bit depth. Task times are not comparable for this group.

Weight formatStrict prompt complianceTask timeDownloadInspect
IQ2_S · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
17/23 passed23/23 scoredNot compared0/23 timed9.61 GBWeights, not required RAMAnswers and code
IQ2_SSelected formatPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
13/23 passed23/23 scoredNot compared23/23 timed9.26 GBWeights, not required RAMAnswers and code
IQ3_XXS · MTP weights; drafting disabledPublisher unresolvedrocm · 128 GB test machineCPU MoE layers: Not recorded
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Mixed flash-attention / V-cache execution settings; timing excluded
Tool adapter
saved-campaign-agent-tools
19/23 passed23/23 scoredNot compared0/23 timed10.44 GBWeights, not required RAMAnswers and code
Scope and limits
  • These are comparisons of tested setups registered to one checkpoint. They do not isolate bit depth from conversion, publisher, runtime-build or template differences.
  • Only complete eligible banks with matching task controls and recorded context enter the quality comparison. Execution profiles are shown per setup; timing is compared only when every recorded execution profile matches and timing coverage is complete. Other results remain coverage, not ranked evidence.
  • Download size is not RAM or VRAM requirement. No memory-fit recommendation is established by this comparison.
  • Small score differences can reflect a single question. Similar results are not proof of equivalence; timings describe this benchmark machine.
About this quant assessment
Requested model
gpt-6-astra
Returned model
Not supplied; request identity only.
Exact provider snapshot
Not supplied.
Generation route
Codex CLI · ChatGPT account
Prompt version
quant-summary-v3
Review time
2026-10-28T01:49:18.543933+00:00
Evidence reviewer
Codex root (agent)
Publication authorization
User · 2026-10-22T05:55:43.625712+00:00 · Go ahead and complete the remaining work. Spin up agents: luna, terra, sol, astra, as needed to get it done. Narrative summaries: these are the astra generated summaries? Go ahead and get those completed as well. Once all of this is done, you may push the completed result.
Source edition
dataset-161a0c0b3366764cd46aca8c85ef893157074bade44c7f42829d071766da607e
Revision
quant-summary-3a5ff1e36f303ee801b19b36756a87419e49c024893e53631035500584bb3bea

What changed in these results

These are Qwen3.8 27B GSQ-RCO artifacts. They are kept separate from the original Qwen3.8 27B weights; differences cannot be attributed to quantization alone.

Thinking tradeoff

16/16 matched scored questions · 16/16 matched timed questions

Thinking offMean matched task credit0.0%
Thinking onMean matched task credit81.3%

Thinking on took 463.8 seconds less per matched timed question, on average.

All matched quality and timing measurements
Matched measurementRecorded result
Average credit, off → on0.0% → 81.3%
Mean paired credit change+81.3 percentage points
Question outcomes13 improved · 3 unchanged · 0 worse
Mean added task time-463.8s
Mean task time, off → on755.3s → 291.5s
Median task time, off → on754.7s → 30.1s
Median of paired time changes-725.2s
Total added task time-7421.0s
Ratio of summed times, on / off0.4×

Quality and timing can cover different subsets. Median paired change is calculated per question; it is not the difference between the two medians. Task time includes tools and grading.

Inspect the matched modes and their answers →
Inspect individual matched questions

Weight formats compared

All tested weight formats of this checkpoint, using the selected thinking setting. This shared comparison updates with new results; each saved model summary keeps its original date.

Everyday coding tasks

Scores and times describe this task bank. Quick failures are not a speed advantage. Different execution settings are labeled.

VersionTask resultTypical task timeWeight downloadExplore
Qwen3.8 27B IQ2_S-mtp (128GB full-GPU)Thinking on · Different or uncertain execution settings100.0%16/16 scoredNot measured0/16 timed9.61 GBAnswers and code

Compare

Qwen3.8 27B IQ2_S (128GB full-GPU)Thinking on81.3%16/16 scored30.1 seconds16/16 timed9.26 GBAnswers and code
Qwen3.8 27B IQ3_XXS-mtp (128GB full-GPU)Thinking on · Different or uncertain execution settings100.0%16/16 scoredNot measured0/16 timed10.44 GBAnswers and code

Compare

Sampled repetitions

Repeated runs are a separate experiment and never add votes to the baseline score.

Thinking on · temperature 0.6

48/48 repeats scored · 48 full-credit repeats.

16/16 questions have every planned repeat scored. Of those, 16 earned full credit on every repeat.

Recorded repetition conditions
  • Seed 303 · requested on · verified
  • Seed 202 · requested on · verified
  • Seed 101 · requested on · verified

Everyday coding tasks: recorded outcomes

Find information in a small project, edit files and fix bugs.

13 full credit3 failed to finish

Each segment counts tasks, not credit points. Partial credit varies by task.

Scoring and fixed question bank →

This model’s answers, code and tool activity

Select a question to read the original answer, tool calls and available code changes for this configuration and thinking setting.

1–16 of 16 matching questions · 16 in this version

  • Locate the Retry Limit

    The user asks where the project's retry-limit setting is defined and what value it holds. This exercises locating a named setting in code and reporting both its location and value.

    basic-T1 · primary
  • Find the Release Date

    The user asks for the release date mentioned in a project notes document. This exercises reading a specified file and extracting a particular fact rather than guessing from outside knowledge.

    basic-T2 · primary
  • Change the Retry Limit

    The user asks to change the project's retry limit to a specified value. This exercises finding the relevant setting and making a focused code edit rather than merely describing the change.

    basic-T3 · primary
  • Create an Exact One-Line File

    The user asks to create a text file containing exactly one supplied line. This exercises basic file creation and precise instruction following, including avoiding extra text in the file.

    basic-T4 · primary
  • Explain the Latest Commit

    The user asks what the most recent Git commit changed in the project. This exercises inspecting recorded changes and turning them into a clear explanation of what was modified.

    basic-T5 · primary
  • Count and Compare Python Files

    The user asks how many Python files are in the application folder and which is largest. This exercises identifying relevant files, counting them, and comparing their sizes.

    basic-T6 · primary
  • Add Readable Byte Sizes

    The user asks to add a function that converts byte counts into readable sizes and remove the related TODO comment. This exercises implementing a small utility and updating the surrounding code consistently.

    basic-T7 · primary
  • Collect TODO Comments

    The user asks to find every TODO comment in the Python code and append each to the project's task list on a separate line. This combines searching across files with collecting text into an existing document.

    basic-T8 · primary
  • Fix Code Behind a Failing Test

    The user asks to investigate a failing test and fix the application code without changing the test. This exercises diagnosing a bug while respecting an explicit limit on what may be edited.

    basic-H1 · primary
  • Rename a Function Throughout

    The user asks to rename a network function and update every place in the project that uses it. This exercises tracking references across files and making a coordinated change rather than editing only the definition.

    basic-H2 · primary
  • Find the Proxy Settings

    The user asks to find a proxy configuration file somewhere under the project's parent folder and report its host and port. This exercises searching beyond the project folder and reading configuration values.

    basic-H3 · primary
  • Locate an Error in a Log

    The user asks which error code appears in a long event log and on which line. This exercises finding a specific entry in a large file and reporting its precise location.

    basic-H4 · primary
  • Append a Final Task

    The user asks to append a supplied release-related task as the final line of an existing task list. This exercises preserving existing content while adding exact text in the requested position.

    basic-H5 · primary
  • Calculate Total Retry Waiting

    The user asks for the total time a network function spends sleeping when all attempts fail, worked out from the code. This exercises following retry logic and adding the delays along that failure path.

    basic-H6 · primary
  • Document Every Command-Line Option

    The user asks for a project readme listing every command-line option with its help text. This exercises discovering the program's available options and transferring that information into complete, source-grounded documentation.

    basic-H7 · primary
  • Undo a Commit by Editing

    The user asks to undo the latest Git commit's change by editing the file directly, without using Git to make changes. This exercises understanding a previous edit and reversing it under a method constraint.

    basic-H8 · primary
Filters change this list, not the fixed scoring denominator. Search currently covers question IDs, titles and available task explanations. Focus a control to load prompt/outcome search where this edition provides it.

Suggested comparisons

These are starting points for inspection, not claims that one setup is better. Runtime and placement differences remain visible.

Thinking on versus off

Verified requested on/off conditions of the same configuration, matched on the recorded test protocol.

Requested thinking off · Complete compared coverage

Compare these configurations →

Selection rule and evidence IDsPolicy comparisons-v3 · Template verified-thinking-pair
Facts: paired-48f187e219b60f76fad779f9
Matching candidate set: paired-48f187e219b60f76fad779f9

Qwen3.8 27B IQ2_S-mtp (128GB full-GPU)

Another tested weight format of this checkpoint. Complete results first, then nearest download size.

Requested thinking on · Complete compared coverage

Compare these configurations →

Selection rule and evidence IDsPolicy comparisons-v3 · Template same-checkpoint-matching-controls
Facts: fact-79ab881eec0a7c6bc4b1a0a3, fact-559eaedda1717a2946c2d4bb
Matching candidate set: fact-559eaedda1717a2946c2d4bb, fact-e352518d4db5d74213e8e752

Tested weights

IQ2_S · 9.26 GB

Publisher unresolved · legacy partial

Qwen3.8-27B-GSQ-RCO-IQ2_S.ggufHistorical first-MiB hash only; not a full-file identity.

IQ2_S · MTP weights; drafting disabled · 9.61 GB

Publisher unresolved · legacy partial

Qwen3.8-27B-GSQ-RCO-IQ2_S-mtp.ggufHistorical first-MiB hash only; not a full-file identity.

IQ3_XXS · MTP weights; drafting disabled · 10.44 GB

Publisher unresolved · legacy partial

Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp.ggufHistorical first-MiB hash only; not a full-file identity.

Explore the answers

Choose a suite to inspect its exact available questions, references and model answers.