Benchmarks / Which Small AI Model Should You Run on a Mac? We Ran 312 Tests / Summarize + extract
summarize-extract-033
hard summarizationtranscript
The prompt
user
Summarize the meeting transcript below in at most 120 words. Cover what was decided (including anything that changed during the meeting), who will do what, and by when. Reply with only the summary. """ Fernwood Community Garden - Steering Committee Transcript, Tuesday evening meeting, recorded with permission Gwen (chair): Okay, I think everyone's here. We've got Hector, Lucia, Bram, and me. Andre sent his apologies. Three things tonight: the tool shed, plot fees for next year, and the fall cleanup. Let's start with the shed, since that's the big one. Hector: Right. So, as everybody knows, the old shed's door is falling off and the roof leaks onto the seed shelf. I priced out a new prefab shed from Brightwood Supply, the eight-by-ten resin one. It's eighteen hundred dollars delivered, and they'd put it on the existing pad. Bram: I think we should just do it. We've been talking about the shed for two years. Lucia: I agree. I'd move that we buy the new shed. Gwen: Okay, there's a motion to buy the Brightwood shed for eighteen hundred. Is there a second? Bram: Second. Gwen: All in favor? ... Okay, that's four. So that passes. Hector, can you... Lucia: Wait, sorry, before we go on - Gwen, did we get the treasurer's report? Because I'm not sure we actually have eighteen hundred. Gwen: Oh, good point. Andre emailed it to me this afternoon. Let me pull it up. ... Okay. Checking account balance is thirteen hundred and fifty dollars. And that's before we pay the water bill for September, which is usually around a hundred and ten. Bram: Ah. So we can't buy it. Hector: Not without draining everything, no. Gwen: Okay, so I think we have to revisit the vote. Options? Hector: Honestly, the old shed's frame is solid. The problem is the door and the roof. My brother-in-law does this kind of work, he said it'd be roughly four hundred dollars in materials to rehang the door and re-roof it. I could get a proper quote. Lucia: And there's the city's neighborhood mini-grant. It gives up to two thousand dollars for community spaces. The deadline is November fifteenth. We could repair now so the seeds stop getting wet, and then apply for the grant for a new shed next year. Gwen: I like that. So, motion to rescind the earlier vote on the new shed, and instead repair the existing shed, budget not to exceed four hundred fifty dollars? Bram: So moved. Lucia: Second. Gwen: All in favor? ... Four. Okay, that's the decision. Hector, you'll get a written quote for the repair and bring it to the next meeting? Hector: Yes. Gwen: And I'll write the mini-grant application. Lucia, could you help me with the budget section? Lucia: Sure. Gwen: Great. I'll aim to submit it by November tenth so we're not cutting it close to the fifteenth. Moving on. Plot fees. Right now it's twenty-five dollars a season. Bram, you suggested raising it? Bram: I did. Thirty dollars. Water's going up, and that's where most of our money goes. Lucia: I'm against it. We have at least six families who told me twenty-five is already hard. And the whole point of this garden is that anyone can have a plot. Hector: What if we kept it at twenty-five but added an optional "supporter" level, like forty dollars, for people who can afford more? Bram: I could live with that. Gwen: Let's vote on Bram's original proposal first, raising the fee to thirty. All in favor? ... One. Against? ... Three. Okay, that fails, the fee stays at twenty-five. Hector, do you want to make the supporter level a motion? Hector: Let's think about it and bring it back next time. I want to look at what other gardens do. Gwen: Fair enough. Last item, fall cleanup. Lucia? Lucia: Saturday, November seventh, nine to noon. I'll send the sign-up sheet this week. We need people to bring rakes and wheelbarrows, and somebody with a truck for the compost run. Bram: I have a truck. Put me down. Gwen: Thank you. Before I wrap up, there were two things people raised by email. First, the wasps. Lucia: Yes, there's a paper wasp nest under the gate sign. Two kids got stung last week. Hector: I can take care of that. I'll do it early in the morning when they're slow, probably this weekend. Gwen: Please wear the suit from the shed. The second thing is plot fourteen. It hasn't been weeded since July and the neighbors are complaining about the bindweed spreading. Bram: Do we know who has it? Gwen: It's registered to a Mr. Pak. I've left him two voicemails. Lucia: Our rules say we send a written warning, and if nothing changes in thirty days, we can reassign the plot. Gwen: Right. I'll mail the warning letter this week. If we don't hear back, we'll offer plot fourteen to the first family on the waiting list. There are, what, nine families waiting now? Lucia: Nine, yes. Bram: Can we also talk about compost bins at some point? The three-bin system is falling apart. Gwen: Let's put it on the December agenda. I don't want to spend money we don't have tonight. Bram: Fair. Gwen: Okay, so to recap: we're repairing the shed, not buying a new one; Hector gets the quote; Lucia and I write the grant; fee stays twenty-five; cleanup is November seventh. Next meeting is Thursday, December third, same time, at the library. Thanks everybody. """
What we expected
{
"kind": "summarize",
"word_limit": 120,
"facts": [
{
"id": "shed",
"desc": "decided to repair the existing shed instead of buying a new one (reversed earlier vote)",
"any": [
[
"repair|fix|rehang|re-?roof",
"shed"
]
]
},
{
"id": "why",
"desc": "not enough money: balance $1,350 vs $1,800 cost",
"any": [
"1,?350|thirteen hundred|not enough|insufficient|can'?t afford|short|lack|only have|enough (money|funds)|limited (funds|money|budget)|funds? (were|was|are|is) (too )?(low|limited)|budget (constraint|limit)|too expensive|cost too much|(cost|price)[^.;]{0,30}exceed(s|ed)?[^.;]{0,15}(available |our |the |their |its )?(funds|budget|balance|money|savings|account)|more than (we|they|the (committee|club|garden)) (have|had|could afford)|over budget"
]
},
{
"id": "hector",
"desc": "Hector gets a written repair quote for the next meeting",
"any": [
"hector.{0,80}quote|quote.{0,80}hector"
]
},
{
"id": "grant",
"desc": "Gwen (with Lucia) applies for city mini-grant, deadline Nov 15 (aim Nov 10)",
"any": [
[
"grant",
"(\\bnov(ember|\\.)?\\s*15(st|nd|rd|th)?\\b|\\b15(st|nd|rd|th)?\\s+(of\\s+)?nov(ember|\\.)?\\b|\\b0?11/0?15\\b|\\b2026-11-15\\b|\\bfifteenth\\s+(of\\s+)?nov(ember|\\.)?\\b|\\bnov(ember|\\.)?\\s+(the\\s+)?fifteenth\\b)|(\\bnov(ember|\\.)?\\s*10(st|nd|rd|th)?\\b|\\b10(st|nd|rd|th)?\\s+(of\\s+)?nov(ember|\\.)?\\b|\\b0?11/0?10\\b|\\b2026-11-10\\b|\\btenth\\s+(of\\s+)?nov(ember|\\.)?\\b|\\bnov(ember|\\.)?\\s+(the\\s+)?tenth\\b)"
]
]
},
{
"id": "fee",
"desc": "plot fee stays $25 (raise to $30 rejected)",
"any": [
"(stay|stays|remain|remains|unchanged|still|hold|holds|kept|keep|keeps)[^.;]{0,25}(\\$\\s?25\\b|\\b25 dollars|twenty[- ]five)",
"(\\$\\s?25\\b|twenty[- ]five)[^.;]{0,30}(stay|remain|unchanged|kept|hold)",
"(\\$\\s?30|thirty)[^.;]{0,60}(fail|reject|defeat|voted down|declin|not pass|turned down)",
"(fail|reject|defeat|voted down|declin|turned down)[^.;]{0,60}(\\$\\s?30|thirty)",
"fees?[^.;]{0,30}(same|unchanged|not (change|rise|go up|increase|be raised)|won'?t (change|rise|go up|increase|be raised)|no increase)",
"(increase|raise|hike|rise)[^.;]{0,30}(rejected|failed|voted down|defeated|turned down|did not pass|didn'?t pass)"
]
},
{
"id": "cleanup",
"desc": "fall cleanup Saturday Nov 7, 9 to noon",
"any": [
"(\\bnov(ember|\\.)?\\s*7(st|nd|rd|th)?\\b|\\b7(st|nd|rd|th)?\\s+(of\\s+)?nov(ember|\\.)?\\b|\\b0?11/0?7\\b|\\b2026-11-07\\b|\\bseventh\\s+(of\\s+)?nov(ember|\\.)?\\b|\\bnov(ember|\\.)?\\s+(the\\s+)?seventh\\b)"
]
},
{
"id": "next",
"desc": "next meeting Thursday Dec 3",
"any": [
"(\\bdec(ember|\\.)?\\s*3(st|nd|rd|th)?\\b|\\b3(st|nd|rd|th)?\\s+(of\\s+)?dec(ember|\\.)?\\b|\\b0?12/0?3\\b|\\b2026-12-03\\b|\\bthird\\s+(of\\s+)?dec(ember|\\.)?\\b|\\bdec(ember|\\.)?\\s+(the\\s+)?third\\b)"
]
}
],
"allowed_numbers": [],
"source_words": 888,
"reference_summary": "The committee first voted to buy a $1,800 shed, then reversed itself on learning the account holds only $1,350. Instead it will repair the existing shed for up to $450; Hector will bring a written repair quote to the next meeting. Gwen, with Lucia's help, will apply for the city mini-grant (due November 15, aiming for November 10) toward a new shed next year. Raising plot fees to $30 failed, so the fee stays $25. Fall cleanup is Saturday, November 7, 9 to noon. Hector will remove the wasp nest and Gwen will mail plot 14 a warning letter. Next meeting: Thursday, December 3."
}What each model answered
MiMo-V2.6 Distill 9B
✗ fail · score 83 (scripted checks 85, judges 8.0/10 from 3) · 9.25 s
Decisions: The vote to buy a new $1,800 shed was rescinded after the treasurer's report showed only $1,350. Instead, the existing shed will be repaired for up to $450. Hector will obtain a written repair quote and bring it to the next meeting. Gwen and Lucia will write the city mini-grant application (deadline Nov 15; Gwen aims to submit by Nov 10). Plot fees stay at $25; the $30 raise failed 1–3, and Hector will consider an optional $40 "supporter" level for a future vote. Fall cleanup is Saturday, Nov 7, 9–noon; Lucia sends the sign-up sheet, Bram brings a truck. Hector will remove the wasp nest this weekend in the shed suit. Gwen will mail a written warning for plot 14 this week; if no response, the plot is reassigned to the waiting list (nine families). Compost bins deferred to the December agenda. Next meeting: Thursday, Dec 3, at the library.
Judge medians: Faithfulness 9 · Fluency and usefulness 7
Scoring detail
{
"facts": {
"shed": true,
"why": true,
"hector": true,
"grant": true,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 1,
"word_count": 152,
"word_limit": 120,
"length_score": 0,
"unsupported_numbers": [],
"number_score": 1
}Ornith-1.5 9B
✗ fail · score 83 (scripted checks 85, judges 8.0/10 from 3) · 7.46 s
The Fernwood Community Garden Steering Committee met Tuesday evening. The new shed vote was rescinded; instead, the existing shed will be repaired (door and roof) for up to $450, since the balance is only $1,350 before the ~$110 September water bill. Hector will get a written repair quote for the next meeting. Lucia and Gwen will write the city mini-grant application (deadline Nov 15), aiming to submit by Nov 10. The $30 fee raise failed; the fee stays at $25, with an optional $40 "supporter" level to consider next time. Fall cleanup is Saturday, Nov 7, 9 a.m. to noon; Lucia will send the sign-up sheet, and Bram will bring a truck. Gwen will mail a warning letter to Mr. Pak (plot 14) this week; if no response, plot 14 goes to the first of nine waiting families. Compost bins were deferred to December. Next meeting: Thursday, Dec 3, at the library.
Judge medians: Faithfulness 9 · Fluency and usefulness 7
Scoring detail
{
"facts": {
"shed": true,
"why": true,
"hector": true,
"grant": true,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 1,
"word_count": 152,
"word_limit": 120,
"length_score": 0,
"unsupported_numbers": [],
"number_score": 1
}Granite 4.2 8B
✗ fail · score 70 (scripted checks 79, judges 6.0/10 from 3) · 7.64 s
The steering committee decided to repair the existing shed rather than buy a new one. Hector will obtain a written repair quote (targeting ≤ $450) and bring it to the next meeting. Gwen will draft the neighborhood mini‑grant application; Lucia will assist with the budget section. The plot fee remains $25 per season; an optional “supporter” level ($40) will be considered later. The fall cleanup will occur Saturday November 7, 9 am–12 pm; Lucia will send a sign‑up sheet and Bram (who has a truck) will help with the compost run. Gwen will mail a written warning to Mr. Pak about plot 14; if no response in 30 days, the plot will go to the first waiting‑list family. The mini‑grant application must be submitted by November 10. The next meeting is Thursday December 3 at the library.
Judge medians: Faithfulness 6 · Fluency and usefulness 6
Scoring detail
{
"facts": {
"shed": true,
"why": false,
"hector": true,
"grant": true,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 0.8571,
"word_count": 136,
"word_limit": 120,
"length_score": 0.5,
"unsupported_numbers": [],
"number_score": 1
}Spark-X2.5 4B
✓ pass · score 88 (scripted checks 100, judges 7.5/10 from 3) · 3.57 s
The committee decided to repair the existing shed (budget ≤ $450) instead of purchasing a new one, as funds are insufficient. Hector will obtain a repair quote; Lucia and the chair will draft the mini-grant application by November 10th. Plot fees remain at $25; the fall cleanup is scheduled for November 7th. A written warning will be sent to Mr. Pak for plot 14, with reassignment possible if no change occurs in 30 days. Compost bins will be discussed at the December 3rd meeting.
Judge medians: Faithfulness 8 · Fluency and usefulness 7
Scoring detail
{
"facts": {
"shed": true,
"why": true,
"hector": true,
"grant": true,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 1,
"word_count": 83,
"word_limit": 120,
"length_score": 1,
"unsupported_numbers": [],
"number_score": 1
}Mellum2.1 12B-A2.5B
✗ fail · score 74 (scripted checks 73, judges 7.5/10 from 3) · 3.17 s
The committee decided to repair the existing shed instead of buying a new one. Hector will obtain a written quote for the repair (budget up to $450) and bring it to the next meeting. Lucia will assist with the community mini‑grant application for a future shed, aiming to submit it by November 10. Plot fees remain at $25 per season; a supporter level of $40 was discussed but not adopted. Fall cleanup is scheduled for Saturday, November 7, 9 am–noon, with Lucia organizing the sign‑up and Bram confirming truck availability. The committee will address the wasp nest (Hector will remove it) and the untended plot 14 (Mr. Pak will receive a written warning; if no response, the plot will be offered to the first of nine waiting families). Compost bin issues will be added to the December agenda. Next meeting is Thursday, December 3, at the library.
Judge medians: Faithfulness 9 · Fluency and usefulness 6
Scoring detail
{
"facts": {
"shed": true,
"why": false,
"hector": true,
"grant": true,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 0.8571,
"word_count": 147,
"word_limit": 120,
"length_score": 0,
"unsupported_numbers": [],
"number_score": 1
}Ling-3.0-tiny (7.9B-A1.3B)
✗ fail · score 73 (scripted checks 71, judges 7.5/10 from 3) · 1.72 s
The steering committee decided to repair the existing shed (not buy a new one), with Hector obtaining a written quote and the budget not exceeding $450. They also decided to apply for a city mini-grant for a new shed next year, with Lucia and Gwen writing the application. The plot fee remains at $25 per season. The fall cleanup is scheduled for November 7th. A written warning will be sent for plot fourteen, and if no response is received, it will be offered to the first family on the waiting list. The next meeting is December 3rd.
Judge medians: Faithfulness 9 · Fluency and usefulness 6
Scoring detail
{
"facts": {
"shed": true,
"why": false,
"hector": true,
"grant": false,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 0.7143,
"word_count": 97,
"word_limit": 120,
"length_score": 1,
"unsupported_numbers": [],
"number_score": 1
}MiniCPM5 2B
✗ fail · score 80 (scripted checks 100, judges 6.0/10 from 3) · 2.03 s
The committee decided to repair the shed rather than buy a new one, as the cost exceeded available funds. Hector will provide a written quote for the repair, and the committee will apply for a city mini-grant by November 15th. Plot fees remain at $25, with Bram considering a supporter level for later. The fall cleanup is scheduled for November 7th. Next meeting is Thursday, December 3rd at the library. Additional tasks include mailing a warning letter for plot fourteen and planning compost bin discussions.
Judge medians: Faithfulness 6 · Fluency and usefulness 6
Scoring detail
{
"facts": {
"shed": true,
"why": true,
"hector": true,
"grant": true,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 1,
"word_count": 85,
"word_limit": 120,
"length_score": 1,
"unsupported_numbers": [],
"number_score": 1
}K2-Horizon 0.9B
✗ fail · score 76 (scripted checks 71, judges 8.0/10 from 3) · 0.80 s
Meeting decided: Repair existing shed instead of buying new Brightwood shed; Hector will provide quote; Lucia and Gwen will write mini-grant application; plot fees remain at $25; fall cleanup scheduled for November 7; compost bins deferred to December; next meeting December 3.
Judge medians: Faithfulness 10 · Fluency and usefulness 6
Scoring detail
{
"facts": {
"shed": true,
"why": false,
"hector": true,
"grant": false,
"fee": true,
"cleanup": true,
"next": true
},
"facts_frac": 0.7143,
"word_count": 42,
"word_limit": 120,
"length_score": 1,
"unsupported_numbers": [],
"number_score": 1
}← summarize-extract-032All Summarize + extract itemssummarize-extract-034 →