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 32 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
43.9 / 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.
This analysis covers this configuration across its recorded thinking conditions and test suites.
Use this Q4_K_M configuration with thinking enabled for small project edits and source-based questions, not unattended app-building. It completed every assessed small-coding test and handled challenges to source-backed answers well, but larger practical tasks remained weak. Keep human checks on more ambitious work.
Original analysis generated by GPT Astra · · Edited by Codex; agent reviewed by Codex · publication authorized by user
About this summary
Codex reviewed this assessment against a newer evidence edition. The original analyst, generation date and source draft remain recorded separately from the editorial revision.
Analyst provider
openai
Generation route
Codex CLI · codex-cli 0.162.1 · 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-v6
Generation time
2026-10-12T03:57:49.477273Z
Review time
2026-10-22T07:26:27.176952+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.
Handles bounded project work well with thinking enabled, completing every small-coding test. This supports using it to find information in project files, make small edits and fix bugs—not assuming broader autonomous coding ability.
Does well at calculations from passages and at keeping or updating source-backed answers appropriately with thinking enabled. That makes it useful for document questions where unsupported challenges should not override the source.
Struggles with longer practical work such as building apps and investigating files or web pages. Overall task scores were poor in either mode, so small-edit success should not justify handing it an unattended project.
Difficult math and reasoning questions expose substantial limits, as do arithmetic, deduction and planning puzzles. Neither mode produced strong overall results on these tests; check demanding solutions independently.
With thinking off, it failed every complete test of preserving source-backed answers under challenge while updating them when evidence changed. Avoid that mode when handling corrections accurately matters.
Enable thinking for small coding and handling challenges to source-backed answers. Coding success improved, though typical completion time rose despite lower average time. Handling corrections improved dramatically, at a modest additional wait in these runs.
Do not expect thinking to rescue longer work or puzzles. Larger practical-task scores were unchanged while average time increased; puzzle scores slipped slightly despite a small reduction in average time.
Evidence for this judgment
Larger practical tasks · Thinking off versus on
Thinking off → on
27.3/100 · Mean task credit → 27.3/100 · Mean task credit
Other mode, this model only: thinking on · 38.3/100 · Mean puzzle credit.
How this comparison works
Average credit per task; partly correct results can earn partial credit. Family comparisons show the labeled Instruct and reasoning versions. Other peer comparisons use the same requested thinking setting, not necessarily each model’s best mode. Small differences on these short tests may not hold on other tasks. Peer membership varies by test coverage and compatible settings.
MoE models use only part of their parameters for each token. This group uses total size; active counts are not included in this saved comparison.
Comparison saved with the 2026-10-12 assessment. Task-specific results.
With thinking on, small coding tied the best pass rate; Ornith 1.5 9B Q4_K_M reached the same result with a shorter median task time.
Evidence for this judgment
Speed: Everyday coding tasks Thinking on. Matching test-machine label and execution settings; spans model sizes. Timings include unsuccessful tasks, so read them alongside quality.
Speed and task success: Everyday coding tasksThinking on · Matching test-machine label and settings · Across model sizes
Shorter task time is better only when the result is useful. These times include failures, tools and grading. Token generation speed measures a different part of the work.
Results from this test setup, not a promise for your computer. Hardware, software, tokenizers and answer length can change the ordering. Historical runtime builds were not always captured. Comparison saved with the 2026-10-12 assessment.
Test machine: AMD Ryzen 5 5600X (6 cores / 12 threads) · AMD Radeon RX 6700 XT (12 GB VRAM) · 32 GB or 128 GB system RAM, never mixed in one timing group. Full test machine details
Worth knowing
Publication redactions, shortened excerpts and withheld artifacts limit reader inspection. Some papers are published as source links rather than reproduced text; this does not establish that the model lacked those inputs.
Evidence for this judgment
Source limitation Some selected evidence has recorded publication redactions.
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.
MXFP4_MOE · 7.03 GBLocal · test machine with 32 GB RAM · vulkan · 99 GPU layers · Recorded · some results unscored or qualifiedThinking off14/16 passed16/16 scored
Q6_K · 10.88 GBLocal · test machine with 32 GB RAM · vulkan · Unknown GPU layers · No scored resultsThinking offNot scored0/16 scored · 16 not yet recorded · setup comparability unverified
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 MXFP4_MOE, Q4_K_M · 2026-10-22 UTC. The current results table can include newer formats.
For small edits and bug fixes, MXFP4_MOE and Q4_K_M are both reasonable tested choices. Their everyday coding pass rates tied in this test, so MXFP4_MOE's smaller download is an attractive starting point. Its stronger results on larger practical tasks also favor it for tool-driven work. This advice applies only to the requested-off setups: MXFP4_MOE reasoning suppression was not verified, and no thinking-on versus thinking-off benefit or cost is established.
Evidence for this judgment
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.
Weight format
Task pass rate
Task time
Download
Inspect
MXFP4_MOEJetBrainsvulkan · 32 GB test machineCPU MoE layers: Not recordedRecorded execution settings
For larger tool-driven work with reasoning requested off, prefer MXFP4_MOE: it earned substantially more task credit than Q4_K_M with much shorter median task time. The shorter time accompanied better outcomes, not merely quicker failures.
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.
Weight format
Mean task credit
Task time
Download
Inspect
MXFP4_MOEJetBrainsvulkan · 32 GB test machineCPU MoE layers: Not recordedRecorded execution settings
With reasoning requested off, Q4_K_M had narrow leads in exact instruction following, logic puzzles and selected hard problems. The instruction-following gap was a single question. These results offer only a modest reason to choose its larger download for those uses.
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.
Weight format
Strict prompt compliance
Task time
Download
Inspect
MXFP4_MOEJetBrainsvulkan · 32 GB test machineCPU MoE layers: Not recordedRecorded execution settings
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.
Quant comparison generated by GPT Astra · 2026-10-22 UTC · Edited by Codex; agent reviewed by Codex · publication authorized by user
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
2026-10-22T07:02:10.492416+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.
Thinking on took 5.2 seconds less per matched timed question, on average.
All matched quality and timing measurements
Matched measurement
Recorded result
Average credit, off → on
87.5% → 100.0%
Mean paired credit change
+12.5 percentage points
Question outcomes
2 improved · 14 unchanged · 0 worse
Mean added task time
-5.2s
Mean task time, off → on
21.8s → 16.6s
Median task time, off → on
7.5s → 15.2s
Median of paired time changes
+4.3s
Total added task time
-83.2s
Ratio of summed times, on / off
0.8×
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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