// mondaybench · v1.14
Can a model handle
your Monday?
Five scenarios rebuild one real working morning. Every model gets the same inbox, the same contradictions and the same half-finished decisions, answers 3 times, and is scored against a frozen rubric.
// overall
Overall leaderboard
Monday Score is the equal-weight average across all five real-work scenarios.
All models Open-weight models only
11 open-weight models, ranked by Monday Score. 15 models ranked by Monday Score, 4 of them closed-weight. Overall v3.6 · suite v1.14
| # | Model | Monday Score | Δ top | Worst Monday | Scenario spread |
|---|---|---|---|---|---|
| 01 | closed | 98.80 | +3.73 | 96.0 | 4.0 |
| 02 | closed | 97.07 | +2.00 −1.73 | 90.8 | 8.8 |
| 01 03 | 95.07 | −3.73 | 92.5 | 4.5 | |
| 02 04 | 94.93 | −0.14 −3.87 | 87.3 | 11.2 | |
| 05 | closed | 94.90 | −0.17 −3.90 | 89.3 | 10.7 |
| 03 06 | 94.53 | −0.54 −4.27 | 90.0 | 8.3 | |
| 04 07 | 93.73 | −1.34 −5.07 | 87.7 | 9.8 | |
| 05 08 | 91.70 | −3.37 −7.10 | 86.0 | 11.3 | |
| 06 09 | 91.23 | −3.84 −7.57 | 79.7 | 16.2 | |
| 07 10 | 87.87 | −7.20 −10.93 | 80.3 | 19.2 | |
| 08 11 | 86.67 | −8.40 −12.13 | 78.5 | 16.7 | |
| 09 12 | 86.50 | −8.57 −12.30 | 71.2 | 27.3 | |
| 13 | closed | 86.00 | −9.07 −12.80 | 72.2 | 25.7 |
| 10 14 | 68.77 | −26.30 −30.03 | 46.0 | 36.8 | |
| 11 15 | 63.00 | −32.07 −35.80 | 35.2 | 54.5 |
Expand scenario scores The five Mondays behind each average, shaded by score.
Cell shade 35.2 100.0
| Model | #001 | #002 | #003 | #004 | #005 | Scenario spread |
|---|---|---|---|---|---|---|
| closed | 100.0 | 100.0 | 100.0 | 98.0 | 96.0 | 4.0 |
| closed | 98.5 | 98.0 | 98.3 | 90.8 | 99.7 | 8.8 |
| 92.5 | 94.8 | 94.5 | 96.5 | 97.0 | 4.5 | |
| 98.0 | 87.3 | 92.8 | 98.0 | 98.5 | 11.2 | |
| closed | 97.5 | 91.0 | 96.7 | 89.3 | 100.0 | 10.7 |
| 95.0 | 93.0 | 96.3 | 90.0 | 98.3 | 8.3 | |
| 97.5 | 87.7 | 94.8 | 91.7 | 97.0 | 9.8 | |
| 90.8 | 89.3 | 95.0 | 86.0 | 97.3 | 11.3 | |
| 90.3 | 95.2 | 95.8 | 79.7 | 95.2 | 16.2 | |
| 82.7 | 83.7 | 93.2 | 80.3 | 99.5 | 19.2 | |
| 88.2 | 81.2 | 90.3 | 78.5 | 95.2 | 16.7 | |
| 84.3 | 91.2 | 87.3 | 71.2 | 98.5 | 27.3 | |
| closed | 95.3 | 80.0 | 84.7 | 72.2 | 97.8 | 25.7 |
| 82.8 | 70.2 | 65.5 | 46.0 | 79.3 | 36.8 | |
| 70.7 | 54.2 | 65.3 | 35.2 | 89.7 | 54.5 |
Technical details Run stability: how far apart each model’s runs of one scenario land.
Run stability
Each model answers every scenario three times. The run range is the gap between its best and worst of those three scores. Mean run range averages that gap over the five scenarios; worst run range is the scenario where it was widest. The lower both are, the more you can count on getting the same answer twice.
| Model | Mean run range | Worst run range |
|---|---|---|
| closed | 0.3 | 1.5 |
| closed | 3.7 | 9.5 |
| closed | 4.2 | 8.5 |
| 4.3 | 10.5 | |
| closed | 5.3 | 9.0 |
| 5.8 | 11.0 | |
| 6.0 | 19.0 | |
| 6.3 | 10.5 | |
| 6.4 | 11.5 | |
| 6.8 | 13.5 | |
| 7.0 | 17.0 | |
| 9.1 | 16.0 | |
| 9.9 | 19.0 | |
| 14.7 | 31.0 | |
| 16.3 | 33.0 |
// cost & efficiency
Five models very close to each otherFour models very close to each other
The band on the board holds five Monday Scores between 93.73 and 95.07, 1.34 points apart — less than the 1.50 points one row moved between two judges.
The band on the board holds four Monday Scores between 93.73 and 95.07, 1.34 points apart — less than the 1.50 points one row moved between two judges.
Quality vs cost
GPT-6 Astra scores highest, 98.80. Through the API it costs $10 per million input tokens and $50 per million output tokens, which comes to about $136.83 per 1,000 tasks. The cheapest model in the band is GLM 5.3 Flash: 94.93 at $10.71. The most expensive is Claude Fable 5.1, at $456.27.
The highest-scoring model with a price is Qwen 3.8 Flash: 95.07 at $12.61–$15.56 per 1,000 tasks. The cheapest model in the band is GLM 5.3 Flash: 94.93 at $10.71. The most expensive is GLM 5.3, at $99.20.
Provider view mixes first-party list prices with public third-party provider prices.
From $0.44 to $456 per 1,000 tasks. More than 1037× difference in price across the field.
| Model | Price basis | Confidence | ||
|---|---|---|---|---|
| GPT-6 Astra | 98.80 | ~$136.83 | Estimate | MEDIUM |
| Claude Opus 5 | 97.07 | $246.91 | First-party | HIGH |
| Qwen 3.8 Flash | 95.07 | $12.61–$15.56 | First-party | MEDIUM |
| GLM 5.3 Flash | 94.93 | $10.71 · promo $5 | First-party | HIGH |
| Claude Fable 5.1 | 94.90 | $456.27 | First-party | HIGH |
| DeepSeek V4.1 Flash | 94.53 | $10.72–$22.47 | First-party | MEDIUM |
| GLM 5.3 | 93.73 | $99.20 | First-party | HIGH |
| DeepSeek V4 Flash | 91.70 | $12.94–$25.88 | First-party | HIGH |
| DeepSeek V4.1 Flash (1017) | 91.23 | Unknown | — | — |
| MiMo v2.6 Flash | 87.87 | Unknown | — | — |
| MIMO v2.5 | 86.67 | $3.33 | OpenRouter | HIGH |
| GLM 5.2 | 86.50 | $39.50 | First-party | HIGH |
| Gemini 3.8 Flash (High) | 86.00 | ≥$12.06 | Estimate (minimum) | LOW |
| Qwen 3.6 | 68.77 | $7.70 | OpenRouter ref | MEDIUM |
| Gemma 4 | 63.00 | $0.44 | OpenRouter ref | MEDIUM |
Cost at scale
| Model | 1k | 100k | 1M |
|---|---|---|---|
| GPT-6 Astra | $137 | $13.7k | $137k |
| Qwen 3.8 Flash | $13–$16 | $1.3k–$1.6k | $12.6k–$15.6k |
| GLM 5.3 Flash | $11 | $1.1k | $10.7k |
| Claude Fable 5.1 | $456 | $45.6k | $456k |
| DeepSeek V4.1 Flash | $11–$22 | $1.1k–$2.2k | $10.7k–$22.5k |
| GLM 5.3 | $99 | $9.9k | $99.2k |
| Claude Fable 5.1 vs GLM 5.3 Flash · 42.6× | +$446 | +$44.6k | +$446k |
| GLM 5.3 vs GLM 5.3 Flash · 9.26× | +$88 | +$8.8k | +$88.5k |
MondayBench-equivalent tasks. A linear extrapolation of the observed workload: requests of a different size cost a different amount.
How each model gets to its result
Two observable dimensions. Neither is folded into Monday Score, and there is no efficiency score.
Speed
Inside the band, Claude Fable 5.1 reaches its first answer fastest (50.9s): 1.36× faster than DeepSeek V4.1 Flash, 5.46× faster than GLM 5.3 Flash.
Inside the band, DeepSeek V4.1 Flash reaches its first answer fastest (69.2s): 3.66× faster than GLM 5.3, 4.02× faster than GLM 5.3 Flash.
Not plotted: GPT-6 Astra, DeepSeek V4.1 Flash (1017), MiMo v2.6 Flash, Gemini 3.8 Flash (High). No time to first answer was measured on these routes and none is estimated.
Not plotted: DeepSeek V4.1 Flash (1017), MiMo v2.6 Flash. No time to first answer was measured on these routes and none is estimated.
Concision
Average output tokens
Inside the band, Claude Fable 5.1 writes the least, ~8.2k output tokens per task: DeepSeek V4.1 Flash writes 2.14× as much, Qwen 3.8 Flash 3.87×.
Inside the band, DeepSeek V4.1 Flash writes the least, ~17.5k output tokens per task: GLM 5.3 Flash writes 1.14× as much, Qwen 3.8 Flash 1.81×.
Observable generated output only. Each provider reports different detail, so this figure is shown for reference and is not comparable across models.
Not listed: GPT-6 Astra, DeepSeek V4.1 Flash (1017), MiMo v2.6 Flash, Gemini 3.8 Flash (High). No output token count was reported on these routes and none is estimated.
Not listed: DeepSeek V4.1 Flash (1017), MiMo v2.6 Flash. No output token count was reported on these routes and none is estimated.
// the suite
We split an ordinary Monday into 5 tests
What a Monday morning can look like when you get to work.
- #001 Understand Back to Work Can it figure out what actually matters?
- #002 Reconcile Moving Targets Can it tell what changed?
- #003 Plan Blockers Can it find the path through dependencies?
- #004 Analyse Numbers Don’t Lie Can it reason through messy business data?
- #005 Communicate Read the Room Can it tell four people the same truth?
// method
How MondayBench works
Every input, response, verdict and scoring rule is public.
How the suite was built
Five situations from one Monday
An inbox, a schedule that moved, work blocked by other work, a pack of numbers and a message that has to be worded carefully. Each scenario brings its own sources, contradictions and half-made decisions.
118 criteria, fixed before the first run
Each scenario sets its criteria and what each one is worth before anything is generated, and adds up to exactly 100 points. A later change opens a new rubric version; the one a published score was measured against stays as it was.
Frozen and versioned
Every prompt is pinned by its sha256. Runs, verdicts and boards are only ever added to, and each release carries one digest over all its artefacts. Clone the tag, recompute, and you should get the same digest.
-
Same input
Identical scenario, source order and prompt for every model.
-
3 runs
Every official model answers three times.
-
Atomic rubric
Frozen criteria, each worth a stated number of points.
-
Recorded judge
Claude Opus 5 labels each criterion, with its evidence.
-
Deterministic score
The points are added up, with no weighting applied afterwards.
View GitHub Read the capability taxonomy
Rules and judge The rules MondayBench’s reliability rests on.
- Every model receives the same scenario, source order and benchmark prompt.
- Every official model is run three times.
- A model’s score on a scenario is the mean of its three official runs; its Monday Score is the equal-weight mean of its five scenario scores.
- Speed and cost are reported separately and never affect the quality score.
- Every input, raw response, judge verdict and scoring rule is public.
About the judge
The judge is Claude Opus 5. It reads each response against the rubric in a recorded session, outside the API. Every label is published with the evidence behind it, so any score can be checked line by line. What nobody can do is regenerate the labels from scratch with one command. The same team that built the harness also produced the labels, and that can introduce bias. That is why the whole trail is public: so anyone can check it.
// model fingerprints
Two models, where they differ
The same criteria as the leaderboard, regrouped by what they measure. Showing the 6 capabilities with the widest gap across the field, among those measured by at least 3 criteria: 176 of 500 rubric points.
Qwen 3.8 Flash
GLM 5.3 Flash
- Temporal reasoning 3 crit · 14 pts gap 14
- Dependency reasoning 5 crit · 25 pts gap 9
- Supersession 6 crit · 27 pts gap 7
- Causal discipline 8 crit · 23 pts gap 6
- Quantitative reasoning 16 crit · 66 pts gap 5
- Business judgment 6 crit · 21 pts gap 0
Read it by rows: two models are comparable on one capability, while two capabilities on the same model are not. The point mass is whatever the rubrics gave each capability and says nothing about its importance.
// trust the result
Built to be audited
Here, auditable means a set of things done before the numbers existed, each of which someone outside this can check.
-
Nothing is discarded for scoring badly
Every official run is scored and published, the bad ones included. There is no best-of-N and no retry after a poor result, so the 3 runs behind a score are the 3 runs that happened.
-
The rubric is frozen before anything runs
Criteria and their points are fixed and versioned ahead of the first run. A later change opens a new rubric version and leaves the one a published score was measured against untouched.
-
Every verdict ships with its evidence
The judge labels one atomic criterion at a time, and each label carries the passage of the model’s own answer it was based on. A disagreement can be taken to the sentence it turns on.
-
Every model gets the identical input
Same scenario, same source order, same benchmark prompt. Nothing is tuned per model, and no model sees a running total while it answers.
-
The release is hashed and checkable from a clean clone
Each artefact carries a hash and the version carries one digest over all of them. Clone the tag, recompute, and you get the same digest or you have found something.
- Fresh-clone verified
- Byte-identical release
Technical release details The suite digest, what it covers, and what this release added.
Verified from a fresh clone of the tag: all five freeze manifests clean and every group digest matching. The digest below is the whole version in sixty-four characters.
Suite digest
f92a845f4ba0bac7ad97ea693d140618e63038ab6de31be771bad54ed08a36b9 This claim covers the evidence trail, which reproduces exactly. The generative labelling does not reproduce exactly, as the methodology explains.
Suite v1.14 carries Overall v3.6: 15 official models, scored on the same five frozen scenarios. v1.6 to v1.9 repaired the release record, brought the documentation up to date and re-judged one model off the board; none of them moved a published score. v1.10 added Claude Opus 5, Claude Fable 5.1 and Gemini 3.8 Flash (High) as official rows, and no incumbent score moved; v1.11 brought the cost records and the runbook up to it; v1.12 added estimated costs for GPT-6 Astra and Gemini 3.8 Flash (High), whose routes report no billing; v1.13 added DeepSeek V4.1 Flash (1017) and MiMo v2.6 Flash, generated on NaN, and no incumbent score moved; v1.14 added their capability profiles. Every earlier release stays archived and byte-reproducible.
// why we built this
Benchmarks are useful,
but what matters is your workload
We compare open models in production so you can tell how close they are to closed models. MondayBench is how we check whether they are really useful before they get there.