Benchmark Almanac

DuetBench

Claude Opus 5 1,035 leads since July 2026
Not in model cards

Agents diagnose and improve customer service agents on simulated deployments with procedures, tools and synthetic traffic. A blinded judge compares responses head to head.

Decagon calls it the first customer service benchmark for agent self-improvement. The Fireworks board ranks 9 models by a Diagnosis Elo.

Made by
Decagon
Released
June 2026
Kind
Customer service
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Opus 5 #10 on the capability index
Last model added
August 2026
Scores refreshed
November 12th, 2026
9
Models scored on its own leaderboard

Claude Opus 5 leads with 1,035 among models released by August 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

Bars start at 940.

  1. 1,035 Claude Opus 5
  2. 1,032 GLM-5.3
  3. 1,022 Claude Sonnet 5
  4. 1,011 Kimi K3
  5. 1,009 GLM-5.2
  6. 998 Claude Opus 4.6
  7. 990 GPT-5.6 Sol
  8. 952 GPT-5.6 Terra
  9. 951 GPT-5.6 Luna

Bars start at 940.

  1. 1 Claude Opus 5 1,035
  2. 2 GLM-5.3 1,032
  3. 3 Claude Sonnet 5 1,022
  4. 4 Kimi K3 1,011
  5. 5 GLM-5.2 1,009
  6. 6 Claude Opus 4.6 998
  7. 7 GPT-5.6 Sol 990
  8. 8 GPT-5.6 Terra 952
  9. 9 GPT-5.6 Luna 951

Source: Official DuetBench leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5 1,035 Jul 2026 Official leaderboard
2 GLM-5.3 1,032 Aug 2026 Official leaderboard
3 Claude Sonnet 5 1,022 Jun 2026 Official leaderboard
4 Kimi K3 1,011 Jul 2026 Official leaderboard
5 GLM-5.2 1,009 Jun 2026 Official leaderboard
6 Claude Opus 4.6 998 Feb 2026 Official leaderboard
7 GPT-5.6 Sol 990 Jul 2026 Official leaderboard
8 GPT-5.6 Terra 952 Jul 2026 Official leaderboard
9 GPT-5.6 Luna 951 Jul 2026 Official leaderboard

The best score, 1,035, costs $3.86 a task

Runs on the cost frontier score higher than any run that's cheaper. GLM-5.2 comes within 25.6 of the best for $0.76 a task, 5 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Z.ai OpenAI Anthropic Cost frontier One model at each effort level
0 250 500 750 1,000 1,250 $0.10 $0.20 $0.50 $1.00 $2.00 $5.00 $10 Cost per task, log scale Claude Opus 5 GLM-5.3 GPT-5.6 Terra GPT-5.6 Luna 0 250 500 750 1,000 1,250 $0.10 $1.00 $10 Cost per task, log scale Claude Opus 5 GLM-5.3

Source: Official DuetBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5 1,035 $3.86 18,597
GLM-5.3 1,032 $0.90 9,437
Claude Sonnet 5 1,022 $1.00 15,471
Kimi K3 1,011 $1.22 12,106
GLM-5.2 1,009 $0.76 19,989
Claude Opus 4.6 998 $3.27 13,027
GPT-5.6 Sol 990 $2.12 7,392
GPT-5.6 Terra 952 $0.55 5,024
GPT-5.6 Luna 951 $0.10 7,820

4 models have held number one on the official leaderboard since its release in June 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic Z.ai
Own leaderboard Not tested since Aug 2026
Claude Opus 5 1,035

Source: Official DuetBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
940 960 980 1,000 1,020 1,040 2026 940 960 980 1,000 1,020 1,040 2026

Source: Official DuetBench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Tool use

All 144 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Gemini 4 Argon 90.7%
  • BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%