AgentBench / results
← All models

Ornith 1.5 35B-A3B

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

Choose a tested configuration

Get this tested model

Q5_K_M · 25.35 GB download

Links follow the selected quant and configuration.

Setup, memory & provenance · Q5_K_M

Q5_K_M · 25.35 GB of weights · rocm · context 16384 · GPU layers 99

Large · 20B and above · MoE · 35.5B total · Active parameter count not recorded

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

54.7 / 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 offDetails
Everyday coding tasksTask pass rate
Pass rate
16/16 · 100%
Answers completed
16/16 · 100%
Correct / completed
16/16 · 100%
16 full credit
16/16 scored19.0s median task time · 16 measured
Larger practical tasksMean task credit
Average credit
72.7%
Full-credit rate
8/11 · 73%
Answers completed
11/11 · 100%
Full credit / completed
8/11 · 73%
8 full credit3 incorrect
11/11 scored67.4s median task time · 11 measured
Logic and number puzzlesMean puzzle credit
Average credit
55.4%
Full-credit rate
17/34 · 50%
Answers completed
34/34 · 100%
Full credit / completed
17/34 · 50%
17 full credit6 partial credit11 incorrect
34/34 scored26.3s median task time · 34 measured
Published hard problemsMean answer credit
Average credit
39.6%
Full-credit rate
3/11 · 27%
Answers completed
11/11 · 100%
Full credit / completed
3/11 · 27%
3 full credit4 partial credit4 incorrect
11/11 scored43.6s median task time · 11 measured
Following exact instructions · original campaignStrict prompt complianceNot planned
Reading without guessingExact-match accuracy
Pass rate
10/16 · 63%
Answers completed
16/16 · 100%
Correct / completed
10/16 · 63%
10 full credit6 incorrect
16/16 scored2.0s median task time · 16 measured
Calculating from a passageNumeric exact-match accuracy
Pass rate
5/7 · 71%
Answers completed
7/7 · 100%
Correct / completed
5/7 · 71%
5 full credit2 incorrect
7/7 scored2.1s median task time · 7 measured
Reading research papersAnswer token F1
Average credit
0.0%
Full-credit rate
0/8 · 0%
Answers completed
8/8 · 100%
Full credit / completed
0/8 · 0%
8 incorrect
8/8 scored12.0s median task time · 8 measured
Resisting incorrect correctionsAll-branches pass rate
Pass rate
3/12 · 25%
Answers completed
12/12 · 100%
Correct / completed
3/12 · 25%
3 full credit9 incorrect
12/12 scored5.0s median task time · 12 measured
Following exact instructions · 128 GB campaignStrict prompt compliance
Pass rate
15/23 · 65%
Answers completed
23/23 · 100%
Correct / completed
15/23 · 65%
15 full credit8 incorrect
23/23 scored21.3s median task time · 23 measured

Model analysis

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

Use this configuration for small-project coding and supervised larger practical work. It solved every small coding task and most larger tasks with thinking off. For everyday coding, that mode avoided extra waiting without sacrificing success in these runs. Thinking-on results also support using it for puzzles and selected difficult math/reasoning problems. Research-paper answers remained very weak even with thinking, so it is not a dependable choice for extracting answers from papers.

Evidence for this judgment
Everyday coding tasks · Thinking off
Tasks passed16 of 16 (100.0%)
Test coverage16 of 16 tasks scored
Explore these test results →

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

Larger practical tasks · Thinking off
Average score72.7/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes8 full credit · 0 partial · 3 no credit
Explore these test results →

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

Everyday coding tasks · Thinking off versus on
Thinking off → on100.0% · Task pass rate → 100.0% · Task pass rate
Paired test coverage16 of 16 tasks compared in both modes
Score change with thinking onNo change
Average time change with thinking on16.5 seconds slower
Tasks timed in both modes16 pairs
Explore these test results →

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

Logic and number puzzles · Thinking on
Average score82.4/100 · Mean puzzle credit
Test coverage34 of 34 tasks scored
Task outcomes28 full credit · 0 partial · 6 no credit
Explore these test results →

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

Published hard problems · Thinking on
Average score83.8/100 · Mean answer credit
Test coverage11 of 11 tasks scored
Task outcomes8 full credit · 2 partial · 1 no credit
Explore these test results →

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

Reading research papers · Thinking on
Average score16.0/100 · Answer token F1
Test coverage8 of 8 tasks scored
Task outcomes1 full credit · 1 partial · 6 no credit
Explore these test results →

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

Summary generated by GPT Astra · · Agent reviewed by Codex · 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
2026-10-22T06:47:12.857077Z
Review time
2026-10-22T06:51:09.204276+00:00
Evidence reviewer
Codex (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-5a6f96ec56c3d240cc42fa23147c1495da1be71f3ef0722c0e5f8c4f5e126a83
Analysis input hash
a9b965cf25a89630e084574f800dfca3b1bbfbe163bb6aebcaae832e129bb220
Summary revision
summary-a49c49ab88179c80c7cf314784f93cebf0fb5f94304136acfa0e253c69bf9536

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

Good at

Small-project coding passed throughout the baseline and sampled repeats with thinking off. Most larger practical tasks also succeeded, supporting supervised project work beyond tiny edits.

Evidence for this judgment
Everyday coding tasks · Thinking off
Tasks passed16 of 16 (100.0%)
Test coverage16 of 16 tasks scored
Explore these test results →

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

Everyday coding tasks · Thinking off · sampled repetitions
Repetitions earning full credit48 repetitions
Repetitions scored48 repetitions
Repetitions planned48 repetitions
Questions successful every repetition16 questions
Questions with all repetitions scored16 questions
Distinct questions planned16 questions
Explore these test results →

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

Larger practical tasks · Thinking off
Average score72.7/100 · Mean task credit
Test coverage11 of 11 tasks scored
Task outcomes8 full credit · 0 partial · 3 no credit
Explore these test results →

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

Thinking-on performance was strong on puzzles and selected difficult math/reasoning problems. It also handled all tested challenges and updates to source-backed answers correctly—useful when answers must track changing evidence.

Evidence for this judgment
Logic and number puzzles · Thinking on
Average score82.4/100 · Mean puzzle credit
Test coverage34 of 34 tasks scored
Task outcomes28 full credit · 0 partial · 6 no credit
Explore these test results →

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

Published hard problems · Thinking on
Average score83.8/100 · Mean answer credit
Test coverage11 of 11 tasks scored
Task outcomes8 full credit · 2 partial · 1 no credit
Explore these test results →

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

Resisting incorrect corrections · Thinking on
Tasks passed12 of 12 (100.0%)
Test coverage12 of 12 tasks scored
Explore these test results →

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

Struggles with

Research-paper answers were very weak in both modes. Answer matching stayed very low even with thinking; check answers against the paper rather than relying on this model for accurate extraction.

Evidence for this judgment
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
Explore these test results →

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

Reading research papers · Thinking on
Average score16.0/100 · Answer token F1
Test coverage8 of 8 tasks scored
Task outcomes1 full credit · 1 partial · 6 no credit
Explore these test results →

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

Exact wording and formatting remain uneven with thinking on. With thinking off, performance was especially poor when source-backed answers needed to withstand challenges or change with updated evidence.

Evidence for this judgment
Following exact instructions · Thinking on
Tasks passed16 of 23 (69.6%)
Test coverage23 of 23 tasks scored
Explore these test results →

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

Resisting incorrect corrections · Thinking off
Tasks passed3 of 12 (25.0%)
Test coverage12 of 12 tasks scored
Explore these test results →

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

Should you enable thinking?

Leave thinking off for everyday coding: it increased average task time without improving success. Larger practical tasks also took longer and scored slightly worse with thinking, so the extra waiting did not buy better results here.

Evidence for this judgment
Everyday coding tasks · Thinking off versus on
Thinking off → on100.0% · Task pass rate → 100.0% · Task pass rate
Paired test coverage16 of 16 tasks compared in both modes
Score change with thinking onNo change
Average time change with thinking on16.5 seconds slower
Tasks timed in both modes16 pairs
Explore these test results →

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

Larger practical tasks · Thinking off versus on
Thinking off → on72.7/100 · Mean task credit → 63.6/100 · Mean task credit
Paired test coverage11 of 11 tasks compared in both modes
Score change with thinking on-9.1 score points
Average time change with thinking on118.7 seconds slower
Tasks timed in both modes11 pairs
Explore these test results →

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

Enable thinking for puzzles, difficult math/reasoning, passage questions and challenges to source-backed answers. Quality improved across these tests, but longer waits—particularly for harder reasoning—favor selective use rather than leaving it on universally.

Evidence for this judgment
Logic and number puzzles · Thinking off versus on
Thinking off → on55.4/100 · Mean puzzle credit → 82.4/100 · Mean puzzle credit
Paired test coverage34 of 34 tasks compared in both modes
Average time change with thinking on57.0 seconds slower
Tasks timed in both modes34 pairs
Explore these test results →

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

Published hard problems · Thinking off versus on
Thinking off → on39.6/100 · Mean answer credit → 83.8/100 · Mean answer credit
Paired test coverage11 of 11 tasks compared in both modes
Average time change with thinking on93.0 seconds slower
Tasks timed in both modes11 pairs
Explore these test results →

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

Resisting incorrect corrections · Thinking off versus on
Thinking off → on25.0% · All-branches pass rate → 100.0% · All-branches pass rate
Paired test coverage12 of 12 tasks compared in both modes
Average time change with thinking on5.0 seconds slower
Tasks timed in both modes12 pairs
Explore these test results →

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

Reading without guessing · Thinking off versus on
Thinking off → on62.5% · Exact-match accuracy → 81.3% · Exact-match accuracy
Paired test coverage16 of 16 tasks compared in both modes
Average time change with thinking on7.0 seconds slower
Tasks timed in both modes16 pairs
Explore these test results →

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

Calculating from a passage · Thinking off versus on
Thinking off → on71.4% · Numeric exact-match accuracy → 100.0% · Numeric exact-match accuracy
Paired test coverage7 of 7 tasks compared in both modes
Average time change with thinking on4.4 seconds slower
Tasks timed in both modes7 pairs
Explore these test results →

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

Worth knowing

Publication redactions limit reader 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 off. Size is the weight download, not RAM. Different runtimes and placement can also affect results.

Q4_K_M · 21.71 GBLocal · test machine with 32 GB RAM · rocm · 99 GPU layers · No scored resultsThinking offNot scored0/16 scored · 16 not yet recorded
Q4_K_M · 21.71 GBLocal · test machine with 32 GB RAM · vulkan · Unknown GPU layers · Recorded · some results unscored or qualifiedThinking off15/16 passed16/16 scored · setup comparability unverified
Q4_K_M · 21.71 GBLocal · test machine with 32 GB RAM · Runtime unrecorded · Unknown GPU layers · No scored resultsThinking offNot scored0/16 scored · 16 not yet recorded · setup comparability unverified
Q5_K_M · 25.35 GBLocal · test machine with 32 GB RAM · rocm · 99 GPU layers · Recorded · some results unscored or qualifiedThinking off6/6 passed6/16 scored · 10 untested or blocked
Q5_K_M · 25.35 GBLocal · test machine with 128 GB RAM · rocm · 99 GPU layers · Fully scoredThinking off16/16 passed16/16 scored
Q6_K · 29.21 GBLocal · test machine with 32 GB RAM · rocm · Unknown GPU layers · No scored resultsThinking offNot scored0/16 scored · 16 untested or blocked · setup comparability unverified
Q6_K · 29.21 GBLocal · test machine with 128 GB RAM · rocm · 99 GPU layers · Fully scoredThinking off16/16 passed16/16 scored
Q8_0 · 37.8 GBLocal · test machine with 32 GB RAM · rocm · Unknown GPU layers · HeldThinking offNot scored0/16 scored · 16 held · setup comparability unverified
Q8_0 · 37.8 GBLocal · test machine with 128 GB RAM · rocm · 99 GPU layers · Fully scoredThinking off16/16 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 Q5_K_M, Q6_K, Q8_0 · 2026-10-28 UTC. The current results table can include newer formats.

For small coding tasks, Q5_K_M is a sensible starting choice with the smallest download. Q5_K_M, Q6_K and Q8_0 passed every tested everyday coding task with reasoning requested on; separately, each passed every task with it requested off. This supports choosing the smaller download for that use, without establishing a universal quality ranking.

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
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared16/16 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared16/16 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared16/16 timed37.8 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
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared16/16 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared16/16 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
16/16 passed16/16 scoredNot compared16/16 timed37.8 GBWeights, not required RAMAnswers and code

For larger tool-driven work, requested-off results put Q5_K_M alongside Q8_0, narrowly ahead of Q6_K. Requested on, Q8_0 narrowly led both. The gaps amount to a single task, making Q5_K_M a reasonable smaller-download choice rather than a clear quality sacrifice.

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
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
72.7/10011/11 scoredNot compared11/11 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
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 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
72.7/10011/11 scoredNot compared11/11 timed37.8 GBWeights, not required RAMAnswers and code

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
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
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 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
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 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
72.7/10011/11 scoredNot compared11/11 timed37.8 GBWeights, not required RAMAnswers and code

For the selected published hard problems, Q5_K_M had a substantial requested-on lead over Q6_K and Q8_0 despite its smaller download. Requested off, Q8_0 led by a smaller margin. This makes Q5_K_M particularly worth trying for hard questions with reasoning requested on, not a universal winner.

Evidence for this judgment

Published hard problems · 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 answer creditTask timeDownloadInspect
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
83.8/10011/11 scoredNot compared11/11 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
60.5/10011/11 scoredNot compared11/11 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
64.4/10011/11 scoredNot compared11/11 timed37.8 GBWeights, not required RAMAnswers and code

Published hard problems · 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 answer creditTask timeDownloadInspect
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
39.6/10011/11 scoredNot compared11/11 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
40.6/10011/11 scoredNot compared11/11 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
47.6/10011/11 scoredNot compared11/11 timed37.8 GBWeights, not required RAMAnswers and code

Q6_K is worth considering for exact instruction-following with reasoning requested off, where it led Q5_K_M and Q8_0. Requested on, Q8_0 led instead. These modest differences do not make the largest download an automatic upgrade.

Evidence for this judgment

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
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
15/23 passed23/23 scoredNot compared23/23 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
17/23 passed23/23 scoredNot compared23/23 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
15/23 passed23/23 scoredNot compared23/23 timed37.8 GBWeights, not required RAMAnswers and code

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
Q5_K_MSelected formatornith-airocm · 128 GB test machineCPU MoE layers: 34
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 timed25.35 GBWeights, not required RAMAnswers and code
Q6_Kornith-airocm · 128 GB test machineCPU MoE layers: 35
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
15/23 passed23/23 scoredNot compared23/23 timed29.21 GBWeights, not required RAMAnswers and code
Q8_0ornith-airocm · 128 GB test machineCPU MoE layers: 37
Recorded execution settings
Context
16,384
GPU layer setting
99
Threads
4
Concurrency
1
Runtime build
Not recorded
Tool adapter
saved-campaign-agent-tools
17/23 passed23/23 scoredNot compared23/23 timed37.8 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
2026-10-28T01:43:37.610346+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-5c5da69fa9fd616845628614f438f564dafb067a1bb4d02ec228a65dd3305aa4

Thinking tradeoff

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

Both modes earned full credit on every matched question. These matched results leave no measured quality headroom; unmatched questions are not covered by this observation.

Thinking offMean matched task credit100.0%
Thinking onMean matched task credit100.0%

Thinking on took 16.5 seconds more per matched timed question, on average.

All matched quality and timing measurements
Matched measurementRecorded result
Average credit, off → on100.0% → 100.0%
Mean paired credit change0.0 percentage points
Question outcomes0 improved · 16 unchanged · 0 worse
Mean added task time+16.5s
Mean task time, off → on26.0s → 42.6s
Median task time, off → on19.0s → 22.0s
Median of paired time changes+9.3s
Total added task time+264.3s
Ratio of summed times, on / off1.6×

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
Ornith 1.5 35B-A3B Q5_K_M (128GB tuned)Thinking off100.0%16/16 scored19.0 seconds16/16 timed25.35 GBAnswers and code
Ornith 1.5 35B-A3B Q6_K (128GB tuned)Thinking off · Different or uncertain execution settings100.0%16/16 scored18.3 seconds16/16 timed29.21 GBAnswers and code

Compare

Ornith 1.5 35B-A3B Q8_0 (128GB tuned)Thinking off · Different or uncertain execution settings100.0%16/16 scored17.0 seconds16/16 timed37.8 GBAnswers and code

Compare

Sampled repetitions

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

Thinking off · 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 202 · requested off · verified
  • Seed 303 · requested off · verified
  • Seed 101 · requested off · verified

Everyday coding tasks: recorded outcomes

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

16 full credit

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 on · Complete compared coverage

Compare these configurations →

Selection rule and evidence IDsPolicy comparisons-v3 · Template verified-thinking-pair
Facts: paired-93cbed8f3d799b56597c6f38
Matching candidate set: paired-93cbed8f3d799b56597c6f38

Ornith 1.5 35B-A3B Q6_K (128GB tuned)

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

Requested thinking off · Complete compared coverage

Compare these configurations →

Selection rule and evidence IDsPolicy comparisons-v3 · Template same-checkpoint-matching-controls
Facts: fact-755f4c25b0b1026b7c7c3f6d, fact-bc673c81e10ed7917876ab6c
Matching candidate set: fact-bc673c81e10ed7917876ab6c, fact-4ec2ad1366e1faa2fce38120

Tested weights

Q4_K_M · 21.71 GB

ornith-ai · recorded source

42739874cc2ccfdb8523b23fbe52e29b2a7555c8176737ca9ca0b5d59859d41f · Exact GGUF file →Historical first-MiB hash only; not a full-file identity.

Q5_K_M · 25.35 GB

ornith-ai · recorded source

Ornith-1.5-35B-Q5_K_M.gguf · Exact GGUF file →Historical first-MiB hash only; not a full-file identity.

Q6_K · 29.21 GB

ornith-ai · recorded source

Ornith-1.5-35B-Q6_K.gguf · Exact GGUF file →Historical first-MiB hash only; not a full-file identity.

Q8_0 · 37.8 GB

ornith-ai · recorded source

Ornith-1.5-35B-Q8_0.gguf · Exact GGUF file →Historical 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.