1 tested weight format · Explore results, then inspect the answers.
UD-Q4_K_XL · 74.16 GB download · 3 files required
Links follow the selected quant and configuration.1 technical / answer-channel errors
UD-Q4_K_XL · 74.16 GB of weights · rocm · context 16384 · GPU layers 99
Large · 20B and above · MoE · 119.0B total · Active parameter count not recorded
Memory fit has not been verified for this configuration.
partial
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.
63–64 / 100
Nine categories, equally weighted · 8/9 eligible
Possible score range with unscored tasks receiving zero to full credit. Incomplete results remain unranked; this is not a confidence interval. Quality only; speed is separate. How the index works →
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.
Each suite has its own metric and fixed task bank. The balanced index weights categories equally; individual scores and evidence remain available below.
| Test suite | Thinking on | Details |
|---|---|---|
| Everyday coding tasksTask pass rate |
| |
| Larger practical tasksMean task credit |
| |
| Logic and number puzzlesMean puzzle credit |
| |
| Published hard problemsMean answer credit |
| |
| Following exact instructions · original campaignStrict prompt compliance | Not planned | |
| Reading without guessingExact-match accuracy |
| |
| Calculating from a passageNumeric exact-match accuracy |
| |
| Reading research papersAnswer token F1 |
| |
| Resisting incorrect correctionsAll-branches pass rate |
| |
| Following exact instructions · 128 GB campaignStrict prompt compliance |
|
This analysis covers this configuration across its recorded thinking conditions and test suites.
Use this model for straightforward passage questions and supervised practical work with tools. Thinking off gives the better accuracy–response-time balance for simple passage questions in these runs. With thinking on, it succeeds at most larger practical tasks, but difficult math and logic puzzles still expose clear limits. Choose thinking by task rather than leaving it enabled automatically, and check important outputs rather than treating completion as proof of correctness.
| Tasks passed | 14 of 16 (87.5%) |
|---|---|
| Test coverage | 16 of 16 tasks scored |
| Tasks timed | 16 tasks |
| Typical time per task (median) | 4.5 seconds |
| Tasks passed | 12 of 16 (75.0%) |
|---|---|
| Test coverage | 16 of 16 tasks scored |
| Tasks timed | 16 tasks |
| Typical time per task (median) | 18.3 seconds |
| Thinking off → on | 87.5% · Exact-match accuracy → 75.0% · Exact-match accuracy |
|---|---|
| Paired test coverage | 16 of 16 tasks compared in both modes |
| Tasks timed in both modes | 16 pairs |
| Average score | 63.6/100 · Mean task credit |
|---|---|
| Test coverage | 11 of 11 tasks scored |
| Task outcomes | 7 full credit · 0 partial · 4 no credit |
| Incorrect completed answers | 4 tasks |
| Failed to finish | 0 tasks |
| Average score | 23.6/100 · Mean answer credit |
|---|---|
| Test coverage | 11 of 11 tasks scored |
| Task outcomes | 2 full credit · 3 partial · 6 no credit |
| Average score | 37.6/100 · Mean puzzle credit |
|---|---|
| Test coverage | 34 of 34 tasks scored |
| Task outcomes | 11 full credit · 5 partial · 18 no credit |
Summary generated by GPT Astra · · Edited by Codex; agent reviewed by Codex root · publication authorized by user
Generation and editorial authorship remain attached to this revision. Changing the default generator does not change this attribution.
Answers straightforward passage questions well with thinking off—a useful option for finding information in text, including tests where the passage may not contain an answer.
| Tasks passed | 14 of 16 (87.5%) |
|---|---|
| Test coverage | 16 of 16 tasks scored |
Handles useful larger practical work with thinking on: most tasks succeeded. The remaining outputs were incorrect rather than unfinished, so checking the result still matters.
| Average score | 63.6/100 · Mean task credit |
|---|---|
| Test coverage | 11 of 11 tasks scored |
| Task outcomes | 7 full credit · 0 partial · 4 no credit |
| Incorrect completed answers | 4 tasks |
| Failed to finish | 0 tasks |
Small project edits are a promising preview: most assessed tasks succeeded in both modes, but some were incorrect or unfinished, and technical errors left gaps. Check each change.
| Tasks passed | 12 of 15 (80.0%) |
|---|---|
| Test coverage | 15 of 16 tasks scored |
| Failed to finish | 1 tasks |
| Incorrect completed answers | 2 tasks |
| Unscored technical errors | 1 tasks |
| Tasks passed | 11 of 15 (73.3%) |
|---|---|
| Test coverage | 15 of 16 tasks scored |
| Unscored technical errors | 1 tasks |
| Failed to finish | 1 tasks |
| Incorrect completed answers | 3 tasks |
Difficult math and reasoning are weak spots. Hard-problem scores were low in both modes, and logic-puzzle results remained poor even with thinking enabled.
| Average score | 23.6/100 · Mean answer credit |
|---|---|
| Test coverage | 11 of 11 tasks scored |
| Task outcomes | 2 full credit · 3 partial · 6 no credit |
| Average score | 41.5/100 · Mean answer credit |
|---|---|
| Test coverage | 11 of 11 tasks scored |
| Task outcomes | 4 full credit · 2 partial · 5 no credit |
| Average score | 37.6/100 · Mean puzzle credit |
|---|---|
| Test coverage | 34 of 34 tasks scored |
| Task outcomes | 11 full credit · 5 partial · 18 no credit |
Do not trust source-backed answers to withstand challenges or updates with thinking off: none of these tests passed all required branches. These were incorrect completed outputs, not failures to finish.
| Tasks passed | 0 of 12 (0.0%) |
|---|---|
| Test coverage | 12 of 12 tasks scored |
| Incorrect completed answers | 12 tasks |
| Failed to finish | 0 tasks |
Enable thinking for calculations from passages and handling challenges or updates to source-backed answers: both improved substantially, at a clear time cost. Strict instruction-following also improved, though less dramatically.
| Thinking off → on | 42.9% · Numeric exact-match accuracy → 85.7% · Numeric exact-match accuracy |
|---|---|
| Paired test coverage | 7 of 7 tasks compared in both modes |
| Tasks timed in both modes | 7 pairs |
| Score change with thinking on | +42.9 score points |
| Average time with thinking off | 4.6 seconds |
| Average time with thinking on | 22.5 seconds |
| Thinking off → on | 0.0% · All-branches pass rate → 75.0% · All-branches pass rate |
|---|---|
| Paired test coverage | 12 of 12 tasks compared in both modes |
| Tasks timed in both modes | 12 pairs |
| Score change with thinking on | +75.0 score points |
| Average time with thinking off | 8.1 seconds |
| Average time with thinking on | 48.6 seconds |
| Thinking off → on | 65.2% · Strict prompt compliance → 78.3% · Strict prompt compliance |
|---|---|
| Paired test coverage | 23 of 23 tasks compared in both modes |
| Tasks timed in both modes | 23 pairs |
| Score change with thinking on | +13.0 score points |
| Average time with thinking off | 81.0 seconds |
| Average time with thinking on | 152.8 seconds |
Leave thinking off for straightforward passage questions: it was quicker and more accurate. Research-paper answer matching barely improved despite substantially longer average time; difficult math and reasoning scored worse with thinking.
| Thinking off → on | 87.5% · Exact-match accuracy → 75.0% · Exact-match accuracy |
|---|---|
| Paired test coverage | 16 of 16 tasks compared in both modes |
| Tasks timed in both modes | 16 pairs |
| Score change with thinking on | -12.5 score points |
| Median time with thinking off | 4.5 seconds |
| Median time with thinking on | 18.3 seconds |
| Thinking off → on | 53.8/100 · Answer token F1 → 55.7/100 · Answer token F1 |
|---|---|
| Paired test coverage | 8 of 8 tasks compared in both modes |
| Tasks timed in both modes | 8 pairs |
| Score change with thinking on | +1.8 score points |
| Average time with thinking off | 30.0 seconds |
| Average time with thinking on | 89.2 seconds |
| Thinking off → on | 41.5/100 · Mean answer credit → 23.6/100 · Mean answer credit |
|---|---|
| Paired test coverage | 11 of 11 tasks compared in both modes |
| Tasks timed in both modes | 11 pairs |
| Score change with thinking on | -17.9 score points |
| Average time with thinking off | 244.4 seconds |
| Average time with thinking on | 284.7 seconds |
Technical errors leave small-project coding in both modes, plus thinking-off larger tasks and puzzles, as incomplete previews. Those unscored errors are separate from wrong answers and failures to finish.
Source limitation
Some suite/condition results are incomplete or ineligible; their observed scores are previews.
| Tasks passed | 12 of 15 (80.0%) |
|---|---|
| Test coverage | 15 of 16 tasks scored |
| Unscored technical errors | 1 tasks |
| Tasks passed | 11 of 15 (73.3%) |
|---|---|
| Test coverage | 15 of 16 tasks scored |
| Unscored technical errors | 1 tasks |
| Average score | 22.2/100 · Mean task credit |
|---|---|
| Test coverage | 9 of 11 tasks scored |
| Task outcomes | 2 full credit · 0 partial · 7 no credit |
| Unscored technical errors | 2 tasks |
| Average score | 25.5/100 · Mean puzzle credit |
|---|---|
| Test coverage | 33 of 34 tasks scored |
| Task outcomes | 7 full credit · 8 partial · 18 no credit |
| Unscored technical errors | 1 tasks |
Some publication material is redacted, limiting what readers can inspect.
Source limitation
Some selected evidence has recorded publication redactions.
One weight format is recorded for this model. Additional tested formats will appear here when available.
14/16 matched scored questions · 14/16 matched timed questions · Preview: coverage is incomplete.
Thinking on took 48.5 seconds more per matched timed question, on average.
| Matched measurement | Recorded result |
|---|---|
| Average credit, off → on | 78.6% → 85.7% |
| Mean paired credit change | +7.1 percentage points |
| Question outcomes | 1 improved · 13 unchanged · 0 worse |
| Mean added task time | +48.5s |
| Mean task time, off → on | 76.4s → 124.8s |
| Median task time, off → on | 51.9s → 84.0s |
| Median of paired time changes | +28.9s |
| Total added task time | +678.4s |
| Ratio of summed times, on / off | 1.6× |
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 →Repeated runs are a separate experiment and never add votes to the baseline score.
47/48 repeats scored · 45 full-credit repeats.
15/16 questions have every planned repeat scored. Of those, 14 earned full credit on every repeat.
Incomplete repetition coverage; unresolved questions are not classified as passing all repeats.
Find information in a small project, edit files and fix bugs.
Each segment counts tasks, not credit points. Partial credit varies by task.
Ranking unavailable: incomplete coverage.
Scoring and fixed question bank →
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
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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThe 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 · primaryThese are starting points for inspection, not claims that one setup is better. Runtime and placement differences remain visible.
Verified requested on/off conditions of the same configuration, matched on the recorded test protocol.
Requested thinking off · Exploratory: incomplete or uncertain evidenceCompare these configurations →
Another checkpoint in the same total-parameter band and architecture category, with matching task controls.
Requested thinking on · Exploratory: incomplete or uncertain evidenceCompare these configurations →
Publisher unresolved · legacy partial
Mistral-Small-4-119B-2603-UD-Q4_K_XL-00001-of-00003.ggufHistorical first-MiB hash only; not a full-file identity.
Mistral-Small-4-119B-2603-UD-Q4_K_XL-00002-of-00003.ggufHistorical first-MiB hash only; not a full-file identity.
Mistral-Small-4-119B-2603-UD-Q4_K_XL-00003-of-00003.ggufHistorical first-MiB hash only; not a full-file identity.
Choose a suite to inspect its exact available questions, references and model answers.