TERMS-Bench

Can the agent negotiate well?

LLM agents negotiate against a fixed stochastic scripted counterpart across commerce and bankroll regimes. SE+ is the fraction of feasible surplus captured; AGR+ the feasible agreement rate. Results by Stanford, displayed with attribution.

SE+ (surplus efficiency)

↑ higher is better

Claude Opus 4.6
Anthropic
SE+ (surplus efficiency): 69.4%
AGR+ (agreement rate): 99.3%

Claude Opus 4.669.4%

GLM 5.1
Z.ai
SE+ (surplus efficiency): 68.6%
AGR+ (agreement rate): 95.1%

ZGLM 5.168.6%

Claude Opus 4.7
Anthropic
SE+ (surplus efficiency): 66.0%
AGR+ (agreement rate): 98.2%

Claude Opus 4.766.0%

Gemma 4 31B
Google
SE+ (surplus efficiency): 64.0%
AGR+ (agreement rate): 99.8%

Gemma 4 31B64.0%

Gemini 3.1 Pro
Google
SE+ (surplus efficiency): 63.9%
AGR+ (agreement rate): 99.7%

Gemini 3.1 Pro63.9%

DeepSeek V4 Pro
DeepSeek
SE+ (surplus efficiency): 61.8%
AGR+ (agreement rate): 97.5%

DeepSeek V4 Pro61.8%

GPT-5.5
OpenAI
SE+ (surplus efficiency): 60.6%
AGR+ (agreement rate): 99.2%

GPT-5.560.6%

Qwen 3.6 Plus
Alibaba
SE+ (surplus efficiency): 60.4%
AGR+ (agreement rate): 98.2%

Qwen 3.6 Plus60.4%

Grok 4.20
xAI
SE+ (surplus efficiency): 60.1%
AGR+ (agreement rate): 99.1%

Grok 4.2060.1%

Kimi K2.6
Moonshot
SE+ (surplus efficiency): 59.7%
AGR+ (agreement rate): 97.1%

Kimi K2.659.7%

Doubao Seed 2.0 Pro
ByteDance
SE+ (surplus efficiency): 52.2%
AGR+ (agreement rate): 99.9%

Doubao Seed 2.0 Pro52.2%

Fixed 0.30
Scripted baseline
SE+ (surplus efficiency): 38.7%
AGR+ (agreement rate): 99.9%

fxFixed 0.3038.7%

Fixed 0.10
Scripted baseline
SE+ (surplus efficiency): 29.0%
AGR+ (agreement rate): 94.5%

fxFixed 0.1029.0%

Fixed 0.01
Scripted baseline
SE+ (surplus efficiency): 27.3%
AGR+ (agreement rate): 92.2%

fxFixed 0.0127.3%

GPT-4o mini
OpenAI
SE+ (surplus efficiency): 18.9%
AGR+ (agreement rate): 52.2%

GPT-4o mini18.9%

AGR+ (agreement rate)

↑ higher is better

Doubao Seed 2.0 Pro
ByteDance
SE+ (surplus efficiency): 52.2%
AGR+ (agreement rate): 99.9%

Doubao Seed 2.0 Pro99.9%

Fixed 0.30
Scripted baseline
SE+ (surplus efficiency): 38.7%
AGR+ (agreement rate): 99.9%

fxFixed 0.3099.9%

Gemma 4 31B
Google
SE+ (surplus efficiency): 64.0%
AGR+ (agreement rate): 99.8%

Gemma 4 31B99.8%

Gemini 3.1 Pro
Google
SE+ (surplus efficiency): 63.9%
AGR+ (agreement rate): 99.7%

Gemini 3.1 Pro99.7%

Claude Opus 4.6
Anthropic
SE+ (surplus efficiency): 69.4%
AGR+ (agreement rate): 99.3%

Claude Opus 4.699.3%

GPT-5.5
OpenAI
SE+ (surplus efficiency): 60.6%
AGR+ (agreement rate): 99.2%

GPT-5.599.2%

Grok 4.20
xAI
SE+ (surplus efficiency): 60.1%
AGR+ (agreement rate): 99.1%

Grok 4.2099.1%

Claude Opus 4.7
Anthropic
SE+ (surplus efficiency): 66.0%
AGR+ (agreement rate): 98.2%

Claude Opus 4.798.2%

Qwen 3.6 Plus
Alibaba
SE+ (surplus efficiency): 60.4%
AGR+ (agreement rate): 98.2%

Qwen 3.6 Plus98.2%

DeepSeek V4 Pro
DeepSeek
SE+ (surplus efficiency): 61.8%
AGR+ (agreement rate): 97.5%

DeepSeek V4 Pro97.5%

Kimi K2.6
Moonshot
SE+ (surplus efficiency): 59.7%
AGR+ (agreement rate): 97.1%

Kimi K2.697.1%

GLM 5.1
Z.ai
SE+ (surplus efficiency): 68.6%
AGR+ (agreement rate): 95.1%

ZGLM 5.195.1%

Fixed 0.10
Scripted baseline
SE+ (surplus efficiency): 29.0%
AGR+ (agreement rate): 94.5%

fxFixed 0.1094.5%

Fixed 0.01
Scripted baseline
SE+ (surplus efficiency): 27.3%
AGR+ (agreement rate): 92.2%

fxFixed 0.0192.2%

GPT-4o mini
OpenAI
SE+ (surplus efficiency): 18.9%
AGR+ (agreement rate): 52.2%

GPT-4o mini52.2%

results as of 2026-08-04 · full results at source ↗