Sonnet 5.5 beats your ticket routing model by 11.9 points at 41% lower cost

New release·Sonnet 5.5 shipped Oct 4, 09:00·Scored 412 cases by 10:02

Your ticket routing runs GPT-5.6 Sol at $2.90 per task and scores 35.8%
Sonnet 5.5 scores 47.7% at $1.70 and now holds 4 of the 8 points on your frontier

Quality: 47.7%▲ +11.9 pts
Cost per task: $1.70▼ 0.6×
Estimated monthly cost at 60,000 tasks: $102k▼ −$72k

Fable 5.1GPT-5.6 SolGrok 4.7Opus 5.5 MaxSonnet 5.5ProductionNew champion50%40%30%$15$9$3

Your frontierTicket routingFrontierSonnet 5.5Opus 5.5Others

What changed

All runs

26 models · 412 cases · avg@3

Model	Quality	Cost / task	p95	Tokens / task	Turns	Frontier
Opus 5.5 Max	57.8%	$13.43	190.8s	218,363	185	Yes
Opus 5.5 Extra High	56.0%	$6.98	110.5s	101,083	109	No
Opus 5.5 High	56.0%	$3.97	72.5s	58,972	68	Yes
Sonnet 5.5 Max	55.4%	$9.60	121.7s	171,204	142	No
Sonnet 5.5 High	53.1%	$3.90	71.7s	72,418	81	Yes
Opus 5.5 Medium	52.5%	$2.90	52.3s	41,560	52	Yes
Fable 5.1 Max	51.8%	$17.30	212.1s	164,065	171	No
Fable 5.1 Extra High	51.6%	$13.00	167.3s	121,347	133	No
Fable 5.1 High	49.2%	$9.10	117.9s	84,471	97	No
Sonnet 5.5 Medium	47.7%	$1.70	36.3s	30,118	44	Yes
Fable 5.1 Medium	46.8%	$7.10	98.3s	65,143	78	No
Grok 4.7 Max	46.3%	$6.00	90.2s	117,907	126	No
Fable 5.1 Low	45.1%	$5.50	79.4s	48,564	63	No
Grok 4.7 High	43.9%	$4.70	72.3s	91,727	101	No
Opus 5.5 Low	43.8%	$1.30	31.2s	20,049	31	Yes
GPT-5.6 Sol Extra High	41.8%	$8.30	138.7s	130,686	152	No
Grok 4.7 Medium	41.6%	$3.50	57.1s	70,380	79	No
Gemini 3.8 Flash High	39.7%	$4.70	74.3s	187,621	164	No
Sonnet 5.5 Low	39.2%	$0.70	20.3s	12,589	24	Yes
GPT-5.6 Sol High	37.8%	$4.40	84.8s	71,085	88	No
Gemini 3.8 Flash Medium	37.4%	$4.10	64.5s	162,074	139	No
Sonnet 5.5 Minimal	35.8%	$0.50	15.7s	8,572	19	Yes
GPT-5.6 Sol Medium	35.8%	$2.90	57.2s	46,902	61	No
Grok 4.7 Low	33.1%	$1.60	29.7s	30,930	41	No
GPT-5.6 Sol Low	31.0%	$1.80	37.8s	29,061	39	No
GPT-5.6 Sol Minimal	24.5%	$0.90	20s	14,078	22	No

Where each model lands

Top right is best on both axes, faded dots are off the frontier

Speed against cost

Latency p95 · cost per task

Sonnet 5.5 Minimal is the cheapest and fastest run, at $0.50 and 15.7s per task.

It matches production's 35.8% at a sixth of the cost. Sonnet 5.5 Medium is the pick, scoring 47.7% at $1.70 and 36.3s, both under production's $2.90 and 57.2s.

Quality against speed

Quality · latency p95

Opus 5.5 High gives the best quality without the wait, 56.0% at a 72.5s p95.

Opus 5.5 Max adds only 1.8 points and takes 118.3s longer. Sonnet 5.5 Medium is twice as fast at 36.3s for 8.3 points less.

Token appetite

Quality · tokens per task

Opus 5.5 High is the most token-efficient top scorer, 56.0% on 58,972 tokens per task.

Gemini 3.8 Flash High burns 187,621 tokens for 39.7%, over three times as many for 16.3 points less. Sonnet 5.5 Medium uses 30,118, about a third fewer than production's 46,902.

Turns against quality

Quality · turns per task

Opus 5.5 Medium needs the fewest turns of any run above 50%, scoring 52.5% in 52 turns.

Sonnet 5.5 Medium takes 44 turns, 17 fewer than production, and scores 11.9 points higher. Opus 5.5 Max needs 185 turns for 57.8%.

How this was scored

Cases: 412; from your support ticket history
Runs: avg@3; 3 runs per case, scores averaged
Grading: Rubric; grader v14
Next run: Next release; Anthropic and OpenAI releases watched

Release history

Date	Release	Champion	Result
Oct 4	Sonnet 5.5	Sonnet 5.5	Better on ticket routing
Sep 9	GPT-5.6 Sol	GPT-5.6 Sol	Better on ticket routing
Aug 12	Opus 5.5	Opus 5.5	Better on refund approvals and reply drafting
Jun 24	Fable 5.1	Opus 5.1	Same on refund approvals at 4× the cost
May 6	GPT-5.5	Opus 5.1	Slower on reply drafting
Feb 18	GPT-5.4	GPT-5.4	Better on ticket routing