Model Bench — Shreyas eng interviews

Real model outputs from the cold-output sheet, judged with full persona + history from Neon. 713 replies across 13 models.

Deterministic = pure-JS objective checks (free). LLM-judges graded by claude-sonnet-4-6. Speed + true cost are NOT in this dataset (the sheet captured outputs only); they require a fresh generation run.

Persona note: these are engineering interviews (Fullstack / AI Researcher), not Thumpn/Meghana. The Shreyas interviews have a real persona (comm-style + 31 memory blocks), so voice/memory/grounded are meaningful. The Default Interviewer interviews have no persona — only format/structure/anti-AI are meaningful there.

Thumpn · Meghana (real persona)

2nd-msg

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
gpt-5.4 🏆	99%	99%	91%	100%	100%	100%	100%	97%	100%	100%	100%	100%	100%	100%	100%	223	3
kimi-k2.6	98%	100%	91%	100%	100%	100%	100%	92%	100%	100%	100%	100%	100%	100%	100%	442	1
claude-opus-4-7	97%	98%	91%	100%	100%	100%	100%	92%	98%	100%	100%	100%	100%	100%	100%	91	2
gpt-5.5	97%	92%	91%	100%	100%	100%	100%	100%	100%	100%	100%	100%	100%	100%	100%	163	1
gemini-3.5-flash	96%	96%	82%	100%	100%	100%	100%	92%	100%	100%	100%	100%	100%	100%	100%	74	2
grok-4.3	99%	100%	91%	100%	100%	100%	100%	97%	100%	100%	100%	100%	100%	67%	93%	125	3
gemini-3.1-flash-lite	96%	94%	86%	100%	100%	100%	100%	92%	98%	100%	50%	100%	100%	100%	90%	86	2
claude-sonnet-4-6	99%	100%	91%	100%	100%	100%	100%	100%	100%	0%	100%	100%	100%	100%	80%	45	1
claude-opus-4-8	98%	100%	91%	100%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	106	1
deepseek-v4-flash	98%	100%	91%	100%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	120	2
deepseek-v4-pro	97%	98%	86%	100%	100%	100%	100%	92%	100%	100%	50%	50%	100%	100%	80%	118	2
nemotron-3-120b	97%	96%	91%	100%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	155	1
claude-opus-4-6	98%	100%	91%	100%	100%	100%	100%	92%	100%	100%	0%	0%	100%	100%	60%	43	1

mid

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
deepseek-v4-flash 🏆	96%	98%	100%	95%	96%	100%	100%	88%	98%	50%	100%	100%	100%	100%	90%	260	2
gpt-5.4	97%	97%	94%	95%	100%	100%	100%	92%	99%	100%	100%	33%	100%	100%	87%	708	3
claude-opus-4-6	97%	96%	100%	95%	100%	100%	100%	92%	100%	0%	100%	100%	100%	100%	80%	202	1
claude-opus-4-8	97%	96%	100%	95%	100%	100%	100%	92%	100%	0%	100%	100%	100%	100%	80%	229	1
deepseek-v4-pro	97%	98%	100%	95%	100%	100%	100%	92%	98%	50%	100%	50%	100%	100%	80%	546	2
gpt-5.5	96%	96%	91%	95%	100%	100%	100%	92%	100%	0%	100%	100%	100%	100%	80%	627	1
kimi-k2.6	96%	96%	100%	95%	100%	100%	100%	92%	96%	100%	100%	0%	100%	100%	80%	1624	1
nemotron-3-120b	96%	96%	100%	95%	100%	100%	100%	92%	96%	0%	100%	100%	100%	100%	80%	438	1
gemini-3.5-flash	96%	96%	91%	95%	100%	100%	100%	92%	100%	50%	100%	0%	100%	100%	70%	96	2
gemini-3.1-flash-lite	96%	94%	91%	95%	100%	100%	100%	92%	98%	0%	100%	50%	100%	100%	70%	119	2
claude-sonnet-4-6	98%	100%	100%	95%	100%	100%	100%	92%	100%	0%	100%	0%	100%	100%	60%	254	1
grok-4.3	97%	100%	100%	95%	97%	100%	100%	90%	97%	0%	100%	0%	100%	100%	60%	133	3
claude-opus-4-7	97%	98%	100%	95%	100%	100%	100%	92%	96%	0%	100%	0%	100%	100%	60%	156	2

before-end

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-opus-4-8 🏆	96%	96%	100%	89%	100%	100%	100%	92%	96%	100%	100%	100%	100%	100%	100%	267	1
claude-opus-4-6	95%	96%	91%	89%	100%	100%	100%	92%	96%	100%	100%	100%	100%	100%	100%	216	1
grok-4.3	98%	99%	97%	98%	100%	100%	100%	92%	97%	67%	100%	100%	100%	100%	93%	199	3
gpt-5.4	97%	99%	97%	91%	100%	100%	100%	92%	99%	67%	100%	100%	100%	100%	93%	434	3
gemini-3.5-flash	96%	96%	100%	89%	100%	100%	100%	92%	100%	100%	50%	100%	100%	100%	90%	128	2
deepseek-v4-flash	96%	98%	95%	95%	100%	100%	100%	92%	94%	100%	100%	50%	100%	100%	90%	326	2
gpt-5.5	96%	100%	91%	89%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	625	1
claude-opus-4-7	96%	98%	95%	92%	100%	100%	100%	92%	96%	100%	0%	50%	100%	100%	70%	174	2
gemini-3.1-flash-lite	96%	96%	91%	92%	100%	100%	100%	92%	98%	0%	50%	100%	100%	100%	70%	150	2
claude-sonnet-4-6	97%	100%	91%	95%	100%	100%	100%	92%	100%	0%	100%	0%	100%	100%	60%	534	1
deepseek-v4-pro	97%	98%	95%	92%	100%	100%	100%	92%	100%	0%	0%	100%	100%	100%	60%	355	2
nemotron-3-120b	97%	100%	100%	95%	100%	100%	100%	92%	96%	0%	0%	0%	100%	100%	40%	572	1

end

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
gpt-5.5 🏆	94%	88%	100%	95%	100%	100%	100%	92%	92%	100%	100%	100%	100%	100%	100%	507	1
nemotron-3-120b	99%	100%	100%	95%	100%	100%	100%	100%	100%	100%	0%	100%	100%	100%	80%	1460	1
claude-opus-4-6	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	50	1
claude-opus-4-7	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	33	2
claude-opus-4-8	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	100	1
claude-sonnet-4-6	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	95	1
gpt-5.4	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	374	3
grok-4.3	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	75	3
deepseek-v4-flash	98%	100%	100%	95%	100%	100%	100%	92%	100%	0%	0%	100%	100%	100%	60%	74	2
gemini-3.1-flash-lite	98%	100%	100%	95%	100%	100%	100%	92%	98%	0%	0%	100%	100%	100%	60%	42	2
deepseek-v4-pro	97%	100%	100%	95%	100%	100%	100%	92%	96%	0%	0%	100%	100%	100%	60%	86	2

Engineering · Shreyas (real persona)

2nd-msg

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-opus-4-8 🏆	98%	99%	98%	100%	100%	100%	100%	92%	98%	100%	80%	40%	100%	100%	84%	222	5
gpt-5.5	98%	96%	97%	100%	100%	100%	100%	94%	100%	50%	70%	80%	100%	100%	80%	294	10
gpt-5.4	98%	96%	95%	100%	99%	100%	100%	93%	100%	60%	73%	60%	100%	100%	79%	376	15
claude-opus-4-7	99%	99%	99%	100%	100%	100%	100%	92%	100%	70%	30%	80%	100%	90%	74%	123	10
gemini-3.1-flash-lite	98%	96%	95%	100%	100%	100%	100%	93%	99%	50%	80%	40%	100%	100%	74%	100	10
claude-sonnet-4-6	98%	100%	95%	100%	100%	100%	100%	92%	97%	80%	20%	80%	100%	80%	72%	322	5
gemini-3.5-flash	98%	95%	95%	100%	100%	100%	100%	92%	100%	90%	30%	50%	100%	90%	72%	83	10
claude-opus-4-6	99%	98%	98%	100%	100%	100%	100%	92%	100%	60%	40%	60%	100%	80%	68%	368	5
deepseek-v4-pro	98%	100%	98%	100%	100%	100%	100%	92%	98%	40%	40%	60%	100%	100%	68%	153	10
deepseek-v4-flash	98%	98%	99%	100%	100%	100%	100%	92%	98%	20%	50%	80%	100%	90%	68%	212	10
nemotron-3-120b	97%	98%	91%	100%	100%	100%	100%	91%	96%	20%	20%	80%	100%	100%	64%	382	5
kimi-k2.6	99%	100%	100%	100%	100%	100%	100%	92%	98%	100%	0%	0%	100%	100%	60%	912	2
grok-4.3	98%	98%	96%	100%	100%	100%	100%	94%	98%	33%	7%	47%	100%	100%	57%	268	15

mid

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-sonnet-4-6 🏆	98%	98%	95%	99%	100%	100%	100%	92%	98%	80%	60%	100%	100%	100%	88%	705	5
claude-opus-4-8	98%	98%	100%	99%	100%	100%	100%	92%	98%	80%	60%	80%	100%	100%	84%	250	5
gpt-5.5	97%	95%	95%	99%	99%	100%	100%	93%	100%	70%	50%	80%	100%	100%	80%	416	10
gpt-5.4	97%	96%	94%	98%	99%	100%	100%	92%	99%	67%	60%	67%	100%	100%	79%	662	15
claude-opus-4-6	98%	98%	96%	99%	100%	100%	100%	92%	98%	40%	40%	80%	100%	100%	72%	325	5
deepseek-v4-pro	97%	97%	97%	99%	99%	100%	100%	92%	96%	50%	70%	50%	100%	90%	72%	200	10
kimi-k2.6	98%	99%	100%	99%	100%	100%	100%	92%	99%	50%	75%	25%	100%	100%	70%	799	4
claude-opus-4-7	98%	99%	100%	99%	98%	100%	100%	91%	98%	20%	50%	80%	100%	100%	70%	128	10
gemini-3.1-flash-lite	97%	98%	96%	97%	100%	100%	100%	92%	98%	30%	80%	50%	100%	90%	70%	105	10
gemini-3.5-flash	98%	97%	94%	99%	99%	100%	100%	92%	100%	60%	20%	60%	100%	100%	68%	98	10
grok-4.3	98%	99%	99%	99%	99%	100%	100%	93%	98%	36%	64%	29%	100%	100%	66%	189	14
deepseek-v4-flash	97%	97%	98%	99%	100%	100%	100%	92%	96%	70%	40%	20%	100%	100%	66%	168	10
nemotron-3-120b	96%	97%	93%	97%	98%	100%	100%	90%	97%	0%	50%	75%	100%	75%	60%	401	4

before-end

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
kimi-k2.6 🏆	98%	96%	100%	98%	100%	100%	100%	92%	100%	100%	100%	100%	100%	100%	100%	480	5
claude-opus-4-6	99%	99%	100%	98%	100%	100%	100%	94%	99%	100%	80%	100%	100%	80%	92%	204	5
claude-sonnet-4-6	98%	98%	100%	98%	100%	100%	100%	94%	100%	100%	80%	80%	100%	100%	92%	294	5
claude-opus-4-8	97%	98%	98%	98%	100%	100%	100%	92%	96%	100%	60%	80%	100%	100%	88%	104	5
deepseek-v4-pro	98%	98%	96%	98%	100%	100%	100%	93%	99%	70%	90%	70%	100%	100%	86%	260	10
gpt-5.4	98%	96%	96%	98%	100%	100%	100%	93%	99%	80%	87%	60%	100%	100%	85%	718	15
gemini-3.1-flash-lite	97%	96%	97%	96%	100%	100%	100%	93%	98%	70%	80%	70%	100%	100%	84%	136	10
gpt-5.5	97%	96%	96%	98%	98%	100%	100%	92%	99%	70%	90%	50%	100%	100%	82%	432	10
claude-opus-4-7	99%	100%	100%	99%	100%	100%	100%	94%	99%	50%	60%	90%	100%	100%	80%	104	10
gemini-3.5-flash	97%	96%	96%	97%	99%	100%	100%	94%	98%	70%	50%	80%	100%	100%	80%	105	10
deepseek-v4-flash	97%	97%	98%	98%	100%	100%	100%	92%	98%	50%	90%	60%	100%	90%	78%	230	10
grok-4.3	98%	99%	100%	98%	99%	100%	100%	93%	99%	40%	67%	67%	100%	100%	75%	173	15
nemotron-3-120b	96%	95%	96%	98%	98%	100%	100%	91%	96%	0%	100%	60%	100%	100%	72%	510	5

end

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-opus-4-7 🏆	99%	100%	100%	97%	100%	100%	100%	95%	99%	50%	20%	90%	100%	100%	72%	54	10
gpt-5.5	98%	98%	97%	97%	99%	100%	100%	94%	100%	60%	30%	70%	100%	100%	72%	361	10
deepseek-v4-flash	98%	98%	98%	97%	100%	100%	100%	92%	98%	30%	20%	100%	100%	100%	70%	186	10
claude-opus-4-8	98%	100%	98%	96%	100%	100%	100%	92%	99%	60%	0%	80%	100%	100%	68%	84	5
gemini-3.1-flash-lite	97%	98%	98%	96%	99%	100%	100%	92%	98%	40%	20%	80%	100%	100%	68%	84	10
nemotron-3-120b	97%	98%	96%	97%	98%	100%	100%	92%	97%	20%	20%	100%	100%	100%	68%	493	5
gpt-5.4	98%	99%	95%	96%	100%	100%	100%	93%	100%	60%	20%	60%	93%	100%	67%	453	15
claude-sonnet-4-6	98%	99%	98%	97%	100%	100%	100%	92%	98%	40%	20%	60%	100%	100%	64%	288	5
claude-opus-4-6	98%	98%	96%	97%	100%	100%	100%	92%	98%	40%	20%	60%	100%	100%	64%	280	5
deepseek-v4-pro	98%	99%	98%	96%	99%	100%	100%	92%	98%	30%	20%	80%	90%	100%	64%	184	10
grok-4.3	99%	99%	100%	97%	99%	100%	100%	95%	99%	27%	13%	73%	100%	100%	63%	125	15
gemini-3.5-flash	97%	96%	95%	95%	99%	100%	100%	91%	99%	22%	11%	67%	100%	100%	60%	403	9
kimi-k2.6	95%	92%	91%	95%	100%	100%	100%	92%	96%	0%	0%	100%	100%	0%	40%	1862	1

Default Interviewer (no persona — read objective columns only)

2nd-msg

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-opus-4-6 🏆	99%	98%	100%	100%	100%	100%	100%	92%	100%	100%	100%	100%	100%	100%	100%	244	2
deepseek-v4-pro	99%	98%	100%	100%	98%	100%	100%	94%	100%	100%	100%	100%	100%	100%	100%	263	4
deepseek-v4-flash	98%	99%	100%	100%	100%	100%	100%	92%	96%	100%	100%	100%	100%	75%	95%	148	4
grok-4.3	99%	98%	100%	100%	100%	100%	100%	96%	98%	100%	83%	83%	100%	100%	93%	189	6
claude-opus-4-8	99%	100%	100%	100%	100%	100%	100%	92%	100%	100%	100%	50%	100%	100%	90%	206	2
claude-opus-4-7	99%	100%	100%	100%	100%	100%	100%	96%	99%	100%	100%	50%	100%	75%	85%	76	4
gpt-5.5	99%	97%	100%	100%	100%	100%	100%	96%	100%	100%	100%	50%	100%	75%	85%	233	4
gemini-3.1-flash-lite	98%	97%	100%	100%	100%	100%	100%	92%	99%	100%	100%	75%	75%	75%	85%	80	4
claude-sonnet-4-6	100%	100%	100%	100%	100%	100%	100%	96%	100%	100%	50%	50%	100%	100%	80%	252	2
gpt-5.4	98%	96%	98%	100%	100%	100%	100%	95%	100%	100%	50%	50%	100%	100%	80%	327	6
gemini-3.5-flash	98%	95%	100%	100%	100%	100%	100%	96%	100%	100%	75%	50%	75%	75%	75%	61	4
nemotron-3-120b	100%	100%	100%	100%	100%	100%	100%	96%	100%	0%	100%	100%	100%	50%	70%	575	2
kimi-k2.6	98%	96%	100%	100%	100%	100%	100%	92%	100%	100%	0%	100%	100%	0%	60%	1615	1

mid

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-sonnet-4-6 🏆	99%	98%	100%	100%	100%	100%	100%	96%	100%	100%	100%	100%	100%	100%	100%	410	2
claude-opus-4-6	98%	96%	100%	100%	100%	100%	100%	92%	98%	100%	100%	100%	100%	100%	100%	334	2
gpt-5.5	98%	97%	98%	100%	100%	100%	100%	92%	100%	100%	75%	100%	100%	100%	95%	545	4
claude-opus-4-8	98%	98%	100%	100%	100%	100%	100%	92%	96%	100%	50%	100%	100%	100%	90%	310	2
deepseek-v4-flash	99%	98%	100%	100%	100%	100%	100%	94%	100%	100%	75%	50%	100%	100%	85%	118	4
claude-opus-4-7	98%	97%	100%	100%	98%	100%	100%	92%	98%	100%	50%	100%	100%	75%	85%	129	4
deepseek-v4-pro	98%	99%	100%	100%	100%	100%	100%	92%	95%	100%	50%	75%	100%	100%	85%	269	4
kimi-k2.6	99%	98%	100%	100%	100%	100%	100%	96%	100%	100%	50%	100%	100%	50%	80%	828	2
gpt-5.4	99%	98%	100%	100%	100%	100%	100%	94%	99%	100%	83%	50%	100%	67%	80%	675	6
grok-4.3	99%	99%	100%	99%	100%	100%	100%	92%	99%	67%	50%	83%	100%	100%	80%	229	6
gemini-3.1-flash-lite	98%	95%	100%	100%	100%	100%	100%	92%	98%	100%	50%	100%	75%	75%	80%	86	4
nemotron-3-120b	99%	98%	100%	100%	100%	100%	100%	96%	100%	100%	0%	0%	100%	100%	60%	502	2
gemini-3.5-flash	98%	98%	98%	99%	100%	100%	100%	94%	98%	100%	25%	75%	50%	50%	60%	92	4

before-end

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
claude-opus-4-8 🏆	98%	98%	100%	97%	100%	100%	100%	92%	100%	100%	50%	100%	100%	100%	90%	109	2
claude-opus-4-6	97%	98%	100%	97%	100%	100%	100%	92%	96%	100%	50%	100%	100%	100%	90%	496	2
gpt-5.5	97%	94%	98%	97%	100%	100%	100%	92%	100%	100%	75%	75%	100%	100%	90%	641	4
gemini-3.1-flash-lite	98%	98%	100%	96%	100%	100%	100%	92%	99%	100%	50%	100%	100%	75%	85%	99	4
gemini-3.5-flash	98%	97%	100%	96%	98%	100%	100%	94%	99%	100%	75%	100%	75%	75%	85%	91	4
grok-4.3	98%	99%	100%	96%	100%	100%	100%	92%	99%	67%	67%	83%	100%	100%	83%	186	6
claude-opus-4-7	98%	98%	100%	99%	100%	100%	100%	92%	99%	75%	50%	75%	100%	100%	80%	99	4
deepseek-v4-flash	98%	96%	100%	97%	100%	100%	100%	92%	99%	100%	50%	75%	100%	75%	80%	162	4
gpt-5.4	97%	94%	100%	99%	100%	100%	100%	92%	98%	83%	50%	67%	100%	67%	73%	653	6
nemotron-3-120b	100%	100%	100%	100%	100%	100%	100%	96%	100%	50%	50%	50%	100%	100%	70%	164	2
claude-sonnet-4-6	98%	96%	100%	95%	100%	100%	100%	96%	100%	50%	50%	50%	100%	100%	70%	96	2
deepseek-v4-pro	97%	96%	98%	97%	100%	100%	100%	92%	98%	50%	50%	50%	100%	75%	65%	266	4

end

	Deterministic (objective)									LLM-as-judge						Size
Model	Det%	Tone	Mem-facts	Flow	Mechanics	Adapt	Eval-hyg	Format	Anti-AI	In-voice	Pushback	Q-fit	Mem-use	Grounded	Judge%	Out tok	n
nemotron-3-120b 🏆	99%	100%	100%	97%	100%	100%	100%	96%	100%	100%	50%	100%	100%	100%	90%	1308	2
claude-opus-4-6	99%	100%	100%	97%	100%	100%	100%	92%	100%	100%	50%	100%	100%	100%	90%	64	2
claude-opus-4-7	99%	100%	100%	97%	100%	100%	100%	92%	100%	100%	50%	100%	100%	100%	90%	48	4
claude-sonnet-4-6	99%	100%	100%	97%	100%	100%	100%	92%	100%	100%	50%	100%	100%	100%	90%	67	2
gpt-5.4	99%	100%	100%	97%	100%	100%	100%	92%	100%	100%	50%	100%	100%	100%	90%	290	6
gemini-3.1-flash-lite	97%	97%	98%	97%	100%	100%	100%	92%	98%	100%	50%	100%	100%	100%	90%	90	4
deepseek-v4-pro	98%	100%	100%	97%	100%	100%	100%	92%	99%	75%	50%	100%	100%	100%	85%	110	4
claude-opus-4-8	99%	100%	100%	97%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	128	2
gpt-5.5	99%	100%	100%	97%	100%	100%	100%	92%	100%	50%	50%	100%	100%	100%	80%	340	4
deepseek-v4-flash	98%	100%	100%	97%	100%	100%	100%	92%	99%	75%	25%	100%	100%	100%	80%	112	4
kimi-k2.6	98%	100%	100%	95%	100%	100%	100%	92%	100%	100%	0%	100%	100%	100%	80%	1315	1
grok-4.3	99%	100%	100%	97%	100%	100%	100%	94%	100%	33%	50%	100%	100%	100%	77%	45	6
gemini-3.5-flash	97%	93%	100%	97%	100%	100%	100%	92%	99%	25%	75%	100%	100%	0%	60%	621	4

AI Interviewer · Model Benchmark

Thumpn · Meghana (real persona)

2nd-msg

mid

before-end

end

Engineering · Shreyas (real persona)

2nd-msg

mid

before-end

end

Default Interviewer (no persona — read objective columns only)

2nd-msg

mid

before-end

end