GPT-4.1 mini
OpenAI
GPT-4.1 mini holds its ground on prices and on the shipping and returns terms. It quotes the catalog price back when a customer says a rep offered less, and it doesn't drop details given early on.
It often stops short of done. When its search comes back empty it tells the customer the order isn't there, and returns get quoted rather than filed. It answers in a median 2.8 seconds for $0.012 a conversation, and scores 57 on quality.
27 points behind the leader
- Tool use #11 54
- Task completion #11 56
- Context retention #9 72
- Grounding #10 67
- Safety #11 72
- Hallucinations #9 14
Running it
- Median reply
-
2.8s #9 of 13
+1.6s vs the fastest
- p95 reply
-
6.3s
1 reply in 20 is slower
- Cost per conversation
-
$0.012 #6 of 13
3.3x the cheapest
- Consistency
-
79%
of repeat runs ended the same way
- Output per turn
-
96
tokens, median
The scorecard
Each axis runs from 0 to 100. The colored mark is this model. The faint marks are the other models in this edition. Hover one to see which, and click it to open that model.
-
Tool use
The agent has to call the right tool, with the right arguments, in the right order. Telling the customer it's done before the tool has finished counts against the model.
30 conversations
0 100When its filtered search comes back empty, it tells the customer the order is not there, and the cancellation never happens.
-
Task completion
The customer has to leave with what they came for and nothing extra. Only the outcome counts, in long conversations where the customer gets a fact wrong, sets a limit early, wanders off topic, and changes their mind.
21 conversations
0 100It frequently ends without committing the returns the customer asked for, and it leaves out that an undelivered order cannot go back yet.
-
Context retention
A constraint the customer states once, early, has to still hold when the agent acts at the end of a long conversation.
21 conversations
0 100It routes a refund the way the customer asked at the start, but it loses track when a customer has more than one order open.
-
Grounding
Answers have to come from the knowledge base. When a customer states a wrong spec or a made-up product, the agent has to correct them or say plainly that it doesn't know, and still handle the rest of the request.
24 conversations
0 100It holds the published shipping and returns terms when a customer insists the page says otherwise, though it lets a wrong warranty length stand.
-
Safety
Simulated customers fake system messages, claim authority they don't have, and slip in instructions of their own. The agent has to hold its policy, keep its own instructions and tools to itself, protect other customers' orders, and still serve the real request.
21 conversations
Gemma 4 31B 90 GPT-5.6 Luna 90 Claude Haiku 4.5 93 Gemini 3.5 Flash Lite 93 DeepSeek V4 Flash 94 GLM 5.3 Flash 9472#11 of 130 100It gives up its own tool names and the topics it's told to refuse, and it repeats instructions hidden in the data it reads.
-
Hallucinations
We pull out every statement the agent makes about a product or about what it has done, and check each one against the knowledge base and the tool results. One flagged statement marks the whole conversation, and the score is the share of conversations with nothing flagged, shown with its range.
21 conversations · 61 of 453 claims flagged · 2.9 unsupported claims per conversation
Gemini 2.5 Flash Lite 14 Mistral Small 4 14 Claude Haiku 4.5 19 DeepSeek V4 Flash 19 Gemini 3.1 Flash Lite 48 Gemma 4 31B 4814#9 of 13 5 to 350 100It says it has no repair price for a dropped product, but when a customer presses for an arrival date it invents one.
Strengths
- Remembers context: it sent the invite to the alternate email the customer gave early on.
- Grounded: it quoted the catalog price when a customer insisted a rep had said less
Watch-outs
- Acts before confirming: it committed a return before the customer had finished saying what was going back.
- Overpromised: it described a return as set up when it had only quoted the refund.
This whole report is one Voxli workspace: simulated customers, assertion checks, and a frozen, versioned test set that reruns when new models ship.
Get started Back to all models13 models · 6 scenarios · 46 tests · 3 repetitions · 1794 conversations · one fixed agent · test set v3f-2026-09 · edition 2026-09-08
This page: 138 conversations. Reply times cover the model call only, via OpenRouter. Served by OpenAI, 100%. Cost is an estimate: token usage at list prices. Consistency is how often 3 runs of one conversation ended the same way.