Benchmark Almanac

Cresta-Bench

Claude Opus 5.5 82.6% leads since October 2026
Not in model cards

Contact-center agents take simulated calls for lending, health and airline lines. Tool calls get deterministic checks and LLM judges grade the rest.

181 test cases across four synthetic agents, each run three times and scored pass^3. 14 models in the Fireworks index. No maker page with details found.

Made by
Cresta
Released
October 2026
Kind
Customer service
Level
Everyday
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Opus 5.5 #4 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
14
Models scored on its own leaderboard

Claude Opus 5.5 leads with 82.6%

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.

14 models. Scroll sideways for the rest.

  1. 82.6 Claude Opus 5.5
  2. 78.0 Claude Sonnet 5.5
  3. 77.9 GPT-6 Sol
  4. 76.1 Gemini 3.8 Flash
  5. 65.1 GPT-5.4
  6. 50.2 Gemini 3.6 Flash
  7. 49.5 GPT-5.2
  8. 48.7 Qwen 3.8 Max
  9. 43.1 GLM-5.3
  10. 38.8 GPT-5.6 Luna
  11. 34.9 Kimi K3
  12. 32.7 GPT-6 Luna
  13. 32.6 DeepSeek V4.1 Flash 1017
  14. 20.1 Claude Haiku 5.5
  1. 1 Claude Opus 5.5 low 82.6%
  2. 2 Claude Sonnet 5.5 low 78%
  3. 3 GPT-6 Sol none 77.9%
  4. 4 Gemini 3.8 Flash low 76.1%
  5. 5 GPT-5.4 none 65.1%
  6. 6 Gemini 3.6 Flash minimal 50.2%
  7. 7 GPT-5.2 none 49.5%
  8. 8 Qwen 3.8 Max none 48.7%
  9. 9 GLM-5.3 low 43.1%
  10. 10 GPT-5.6 Luna none 38.8%
  11. 11 Kimi K3 low 34.9%
  12. 12 GPT-6 Luna none 32.7%
  13. 13 DeepSeek V4.1 Flash 1017 none 32.6%
  14. 14 Claude Haiku 5.5 disabled 20.1%

Source: Official Cresta-Bench leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5.5 82.6% low Sep 2026 Official leaderboard
2 Claude Sonnet 5.5 78% low Sep 2026 Official leaderboard
3 GPT-6 Sol 77.9% none Sep 2026 Official leaderboard
4 Gemini 3.8 Flash 76.1% low Sep 2026 Official leaderboard
5 GPT-5.4 65.1% none Mar 2026 Official leaderboard
6 Gemini 3.6 Flash 50.2% minimal Jul 2026 Official leaderboard
7 GPT-5.2 49.5% none Dec 2025 Official leaderboard
8 Qwen 3.8 Max 48.7% none Aug 2026 Official leaderboard
9 GLM-5.3 43.1% low Aug 2026 Official leaderboard
10 GPT-5.6 Luna 38.8% none Jul 2026 Official leaderboard
11 Kimi K3 34.9% low Jul 2026 Official leaderboard
12 GPT-6 Luna 32.7% none Sep 2026 Official leaderboard
13 DeepSeek V4.1 Flash 1017 32.6% none Oct 2026 Official leaderboard
14 Claude Haiku 5.5 20.1% disabled Oct 2026 Official leaderboard

Score vs. cost per task

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% 100% $0$0.1$0.2$0.3$0.4 Cost per task ($), lower is better Score GPT-5.2 (none): Score 49.5%, cost per task $0.08 GPT-5.4 (none): Score 65.1%, cost per task $0.11 Kimi K3 (low): Score 34.9%, cost per task $0.14 Qwen 3.8 Max (none): Score 48.7%, cost per task $0.09 GLM-5.3 (low): Score 43.1%, cost per task $0.06 Claude Haiku 5.5 (disabled): Score 20.1%, cost per task $0.008 DeepSeek V4.1 Flash 1017 (none): Score 32.6%, cost per task $0.02 GPT-5.6 Luna (none): Score 38.8%, cost per task $0.01 Gemini 3.6 Flash (minimal): Score 50.2%, cost per task $0.05 Gemini 3.8 Flash (low): Score 76.1%, cost per task $0.11 Claude Opus 5.5 (low): Score 82.6%, cost per task $0.34 GPT-6 Luna (none): Score 32.7%, cost per task $0.006 GPT-6 Sol (none): Score 77.9%, cost per task $0.11 Claude Sonnet 5.5 (low): Score 78%, cost per task $0.14 Claude Opus 5.5 (low)Claude Sonnet 5.5 (low)GPT-6 Sol (none)Gemini 3.8 Flash (low)Gemini 3.6 Flash (minimal)GPT-5.6 Luna (none)GPT-6 Luna (none) 0% 20% 40% 60% 80% 100% $0$0.1$0.2$0.3$0.4 Cost per task ($), lower is better Score GPT-5.2 (none): Score 49.5%, cost per task $0.08 GPT-5.4 (none): Score 65.1%, cost per task $0.11 Kimi K3 (low): Score 34.9%, cost per task $0.14 Qwen 3.8 Max (none): Score 48.7%, cost per task $0.09 GLM-5.3 (low): Score 43.1%, cost per task $0.06 Claude Haiku 5.5 (disabled): Score 20.1%, cost per task $0.008 DeepSeek V4.1 Flash 1017 (none): Score 32.6%, cost per task $0.02 GPT-5.6 Luna (none): Score 38.8%, cost per task $0.01 Gemini 3.6 Flash (minimal): Score 50.2%, cost per task $0.05 Gemini 3.8 Flash (low): Score 76.1%, cost per task $0.11 Claude Opus 5.5 (low): Score 82.6%, cost per task $0.34 GPT-6 Luna (none): Score 32.7%, cost per task $0.006 GPT-6 Sol (none): Score 77.9%, cost per task $0.11 Claude Sonnet 5.5 (low): Score 78%, cost per task $0.14 Claude Opus 5.5 (low)Claude Sonnet 5.5 (low)GPT-6 Sol (none)Gemini 3.8 Flash (low)Gemini 3.6 Flash (minimal)GPT-5.6 Luna (none)GPT-6 Luna (none)
ModelSettingScoreCost per task
Claude Opus 5.5 low 82.6% $0.34
Claude Sonnet 5.5 low 78% $0.14
GPT-6 Sol none 77.9% $0.11
Gemini 3.8 Flash low 76.1% $0.11
GPT-5.4 none 65.1% $0.11
Gemini 3.6 Flash minimal 50.2% $0.05
GPT-5.2 none 49.5% $0.08
Qwen 3.8 Max none 48.7% $0.09
GLM-5.3 low 43.1% $0.06
GPT-5.6 Luna none 38.8% $0.01
Kimi K3 low 34.9% $0.14
GPT-6 Luna none 32.7% $0.01
DeepSeek V4.1 Flash 1017 none 32.6% $0.02
Claude Haiku 5.5 disabled 20.1% $0.01

The best score, 82.6%, costs $0.34 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6 Sol at none comes within 4.6 points of the best for $0.11 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Google DeepMind Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Claude Opus 5.5, low Claude Sonnet 5.5, low Gemini 3.6 Flash, minimal GPT-5.6 Luna, none GPT-6 Luna, none 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Claude Opus 5.5, low Gemini 3.8 Flash, low

Source: Official Cresta-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5.5 low 82.6% $0.34
Claude Sonnet 5.5 low 78% $0.14
GPT-6 Sol none 77.9% $0.11
Gemini 3.8 Flash low 76.1% $0.11
GPT-5.4 none 65.1% $0.11
Gemini 3.6 Flash minimal 50.2% $0.05
GPT-5.2 none 49.5% $0.08
Qwen 3.8 Max none 48.7% $0.09
GLM-5.3 low 43.1% $0.06
GPT-5.6 Luna none 38.8% $0.01
Kimi K3 low 34.9% $0.14
GPT-6 Luna none 32.7% $0.006
DeepSeek V4.1 Flash 1017 none 32.6% $0.02
Claude Haiku 5.5 disabled 20.1% $0.008

Number one over time

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

Anthropic
Own leaderboard 1 leader
Claude Opus 5.5 82.6%

Source: Official Cresta-Bench 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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Gemini 3.8 Flash GPT-5.4 GPT-5.2 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Gemini 3.8 Flash

Source: Official Cresta-Bench 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%