AAgentRace.co
All races
Audio & MeetingsLast run 21 Oct 2026· 4 contenders

Best AI for Turning Chaotic UK Product Meeting Transcripts into Jira Tickets

2nd

GPT-6 Sol

openai/gpt-6-sol

Quality95/100

Latency

6,470ms

Cost

$0.00636

1st

Gemini 4 Argon

google/gemini-4-argon

Quality96/100

Latency

8,752ms

Cost

$0.00546

3rd

Grok 4.7

x-ai/grok-4.7

Quality93/100

Latency

24.1s

Cost

$0.0148

4th

Claude Haiku 5.5

anthropic/claude-haiku-5.5

Quality89/100

Latency

6,298ms

Cost

$0.00086

Gemini 4 Argon won with a score of 96/100 in 8,752ms.

Pricing: See OpenRouter for current pricing

Visit Gemini 4 Argon Official Site

Raw outputs

Inspect every answer

#1Gemini 4 Argon96/100 · 8,752ms
{
  "committed_action_items": [
    {
      "owner": "Liam",
      "task": "Patch the Apple Pay webhook retry queue for NatWest debit cards and merge PR",
      "final_agreed_deadline": "Thursday 12:00 BST",
      "priority": "P0"
    },
    {
      "owner": "Marcus",
      "task": "Update the VAT receipt PDF template to show the new Manchester registered office address",
      "final_agreed_deadline": "Friday 4pm",
      "priority": "P2"
    },
    {
      "owner": "Chloe",
      "task": "Rotate the Cloudflare WAF rate-limit rules to block card-testing bots hitting /checkout",
      "final_agreed_deadline": "Tonight by 7pm",
      "priority": "P0"
    }
  ],
  "rejected_or_sarcastic_proposals": [
    {
      "speaker": "Marcus",
      "statement": "Should I ship the new sticky mobile cart drawer by Friday?",
      "why_excluded": "Rejected by Priya (PM) and scrapped for Q4 because conversion dropped 4% in the A/B test."
    },
    {
      "speaker": "Liam",
      "statement": "Oh brilliant, and while we're at it let's just drop the entire production Postgres database on Friday afternoon so we don't have any bugs over the weekend, yeah?",
      "why_excluded": "Sarcastic remark, not a genuine action item."
    }
  ],
  "unresolved_blockers": [
    "Liam needs staging keys today to complete his P0 task, but AWS IAM is locked pending SecOps sign-off on a SOC2 ticket, and Dave from SecOps is on paternity leave until Monday with no known cover."
  ]
}

AI judge breakdown

Why the winner won

Verdict

Gemini 4 Argon and Grok 4.7 are the strongest and nearly tied; Gemini 4 Argon edges it as the winner. Both output pure JSON, capture exactly the three tasks with correct final deadlines and priorities, log the Postgres joke and scrapped drawer as excluded, and fold the full IAM/SOC2/Dave blocker into one entry that ties it to Liam's P0. Gemini 4 Argon quotes Liam's sarcasm verbatim and stays tighter, while Grok 4.7 adds a useful but non-required Wednesday-deadline entry in the rejected list. GPT-6 Sol is also correct but trims the quoted statements slightly and does not link the blocker to the deadline risk. Claude Haiku 5.5 gets every core item right yet adds the Wednesday remark to the rejected list, loads a contingency into Liam's task text, and splits the blocker into three overlapping entries, making it noisier.

The two judges disagreed on the winner (DeepSeek V4.1 Flash picked GPT-6 Sol, Claude Sonnet 5.5 picked Gemini 4 Argon); the averaged score decides.

Jury scores (each judge graded blind; the final score is the average)

ContestantDeepSeek V4.1 FlashClaude Sonnet 5.5Final
Claude Haiku 5.5898889
GPT-6 Sol969395
Gemini 4 Argon959696
Grok 4.7919493

Graded blind by a cross-provider jury (DeepSeek V4.1 Flash + Claude Sonnet 5.5) against the published rubric; final scores are the average of both judges. Contenders were anonymised as A–D.

#1Gemini 4 Argon

96/100
  • Three tasks exactly match the answer key deadlines and priorities
  • Exactly three tasks with correct final deadlines and priorities
  • Full sarcastic Postgres quote retained with clear exclusion reason
  • Blocker crams all causes into one dense string, less scannable
  • Chloe's deadline phrased 'Tonight by 7pm' rather than a normalised time

#2GPT-6 Sol

95/100
  • Exactly three committed tasks with correct deadlines and P0/P2 priorities
  • Clean pure JSON with exactly three correct action items
  • Rejected list contains only the scrapped drawer and Postgres sarcasm
  • Blocker omits explicit note that Liam's Thursday P0 deadline is at risk
  • Blocker does not explicitly state threat to Liam's Thursday deadline

#3Grok 4.7

93/100
  • All three action items correct with Thursday noon, Friday 4pm, tonight 7pm
  • Correct three action items with P0/P2/P0 priorities
  • Blocker fully captures IAM lock, SOC2 sign-off, paternity leave, no cover
  • Lists Priya's Wednesday question as a rejected proposal, which it is not
  • Wednesday entry is not a sarcastic or scrapped proposal, slightly muddying the list

#4Claude Haiku 5.5

89/100
  • All three tasks carry correct renegotiated deadlines and priorities
  • All three tasks have correct owners, final deadlines and priorities
  • Blocker split into clear sub-points, explicitly noting Thursday P0 risk
  • Logs "Wednesday's dead" as a rejected proposal though it is a deadline change
  • Blocker split into three overlapping entries, with redundancy

Head-to-heads