Extract JSON · 5 checks · max 400 tokens · benched 2026-11-05

Muse Spark 1.3 vs Muse Glimmer 30B vs Llama 4 Maverick vs Llama 4 Scout on event announcement to structured json

Meta models side by side on "Event announcement to structured JSON": Llama 4 Scout scores 10/10; Llama 4 Scout is the cheapest answer scoring 8+ at $0.03 per 1,000 runs. Outputs, checks, judge reasons, latency and cost.

The prompt every model received

System

You are a data extraction engine. Reply with JSON only, no prose, no code fences.

User

Return one JSON object with keys title (string), date (YYYY-MM-DD), start_time (24-hour HH:MM), location (string), attendees (integer). Use only facts stated in the announcement.

Announcement: Join us for the Autumn Data Meetup on Tuesday 3 November 2026. Doors open at 2 pm and talks start at 2:30 pm at The Loft, 14 Harbour Street. We have confirmed 48 attendees so far and can take up to 60.

Rubric for the judge: Correctly picks the talk start (14:30, not doors), ISO date, integer attendees 48, and a sensible title and location.

Side by side

Every cell is one OpenRouter call at temperature 0 with the prompt's token cap and reasoning effort "low" where the model supports it. Cost is usage × the catalogue rate in models.json. Quality is one judge call to anthropic/claude-haiku-4.5 against the prompt's rubric, cached per prompt version.

Muse Spark 1.3

meta/muse-spark-1.3

10/10

Latency: 1.1s
Cost: $0.00375
Per 1,000: $3.75

145 in · 840 out (742 reasoning) · 16 words · checks 5/5

json_valid
json_keys
contains
contains
regex

{"title": "Autumn Data Meetup", "date": "2026-11-03", "start_time": "14:30", "location": "The Loft, 14 Harbour Street", "attendees": 48}

Judge: All fields correctly extracted: title matches announcement, date is proper ISO format, start_time correctly uses talk start time (14:30) not door opening, location is complete and accurate, and attendees uses confirmed count (48) as instructed.

Muse Glimmer 30B

meta/muse-glimmer-30b

10/10

Latency: 113ms
Cost: $0.00028
Per 1,000: $0.28

147 in · 219 out (144 reasoning) · 16 words · checks 5/5

json_valid
json_keys
contains
contains
regex

{"title": "Autumn Data Meetup", "date": "2026-11-03", "start_time": "14:30", "location": "The Loft, 14 Harbour Street", "attendees": 48}

Llama 4 Maverick

meta-llama/llama-4-maverick

10/10

Latency: 399ms
Cost: $6.2e-5
Per 1,000: $0.06

135 in · 51 out · 16 words · checks 5/5

json_valid
json_keys
contains
contains
regex

{"title": "Autumn Data Meetup", "date": "2026-11-03", "start_time": "14:30", "location": "The Loft, 14 Harbour Street", "attendees": 48}

Llama 4 Scout

meta-llama/llama-4-scout

10/10

top judge scorecheapest scoring 8+

Latency: 419ms
Cost: $3.1e-5
Per 1,000: $0.03

135 in · 60 out · 20 words · checks 5/5

json_valid
json_keys
contains
contains
regex

```
{
  "title": "Autumn Data Meetup",
  "date": "2026-11-03",
  "start_time": "14:30",
  "location": "The Loft, 14 Harbour Street",
  "attendees": 48
}
```

Judge: All fields correctly extracted: title matches announcement, date in ISO format (2026-11-03), start_time correctly uses talk start time (14:30 not door opening), location fully specified, and attendees set to confirmed count (48).

Frequently asked

What does this prompt test?

Extract JSON: Correctly picks the talk start (14:30, not doors), ISO date, integer attendees 48, and a sensible title and location. The deterministic checks are json_valid, json_keys, contains, contains, regex.

Which model should I pick for this task?

If the judge's bar of 8/10 is good enough for you, Llama 4 Scout at $0.03 per 1,000 runs. If you need the top score, Llama 4 Scout at $0.03 per 1,000 runs.

Same prompt, other families

Run it yourself

Re-run this prompt on Meta with your own OpenRouter key, or tweak the wording and see what changes.

Partner picks

Claude (credits)Replit Make Taskade Fireflies.ai Pressmaster.ai Murf AI

Affiliate/referral links: commissions or credits

Side by side

Muse Spark 1.3

Muse Glimmer 30B

Llama 4 Maverick

Llama 4 Scout

Frequently asked

Same prompt, other families

Stop overpaying for AI tools