Independent benchmark · EQ Lab

EQ Lab

How local and cloud models handle emotionally complex conversations.

53 models24 personas1187/1272 conversations3721/4149 comparisonsUpdated 8 Nov, 05:20 UTC

Independent experiment, not official EQ-Bench scores. Partner: anthropic/claude-haiku-5.5 via OpenRouter. Method & limitations

Find a model

Results completeProvisional ratingAwaiting judgments

Ability heatmap · 1–10, lower higher relative performance. Missing scores are —. Select a column to sort.

Model ratings and six ability scores
#	Model	Rating	Rapport	Authenticity	Attunement	Needs	Sensemaking	Management
1	GPT-6 Astra (subscription) frontier · subscription	1298.6 95% interval: 1282.6–1312.4 Opus 1264.2 · Sol 1281.4	10.0	10.0	10.0	10.0	10.0	10.0
2	DeepSeek V4.1 Flash 1017 (API, Morph FP8) cloud	1229.2 95% interval: 1209.9–1247.5 Opus 1273.0 · Sol 1165.9	9.4	9.3	8.7	8.1	10.0	8.6
3	Claude Haiku 5.5 (API, Anthropic) cloud	1168.8 95% interval: 1148.2–1191.7 Opus 1184.5 · Sol 1135.5	8.1	8.3	7.8	7.6	9.2	8.0
4	GPT-5.6 Luna (API) cloud	1168.6 95% interval: 1152.5–1185.8 EQ-Bench: 1156.3 Opus 1143.0 · Sol 1167.1	8.2	8.2	7.9	7.9	8.3	8.3
5	Mistral Large 4 (API, Mistral) cloud	1155.3 95% interval: 1132.4–1176.2 Opus 1174.8 · Sol 1123.1	8.4	8.3	7.7	7.3	8.0	7.9
6	Qwen3.8 Flash (API, Alibaba; precision undisclosed) cloud	1154.6 95% interval: 1134.1–1179.7 Opus 1199.8 · Sol 1106.4	8.5	8.0	7.7	6.9	9.2	7.7
7	Qwen3.8 27B (API) cloud	1138.4 95% interval: 1119.5–1154.5 Opus 1163.3 · Sol 1106.7	8.2	7.9	7.6	6.9	8.1	7.6
8	Gemini 3.8 Flash (API, Google AI Studio) cloud	1135.9 95% interval: 1117.3–1154.4 Opus 1171.9 · Sol 1097.5	8.3	7.8	7.6	6.9	8.2	7.4
9	Nemotron 3.5 Super VL 120B-A12B (API, Prime Intellect NVFP4) cloud	1117.1 95% interval: 1096.6–1137.5 Opus 1130.9 · Sol 1097.5	8.0	8.2	7.3	6.3	7.3	7.0
10	Ternary Bonsai 2 27B PTQ1_0 local · thinking on	1110.6 95% interval: 1092.0–1130.8 Opus 1127.3 · Sol 1090.4	7.8	7.7	7.2	6.5	7.4	7.1
11	Muse Glimmer 30B (API, DeepInfra BF16) cloud	1107.9 95% interval: 1089.1–1123.7 Opus 1116.0 · Sol 1094.8	8.0	7.6	7.2	6.3	7.2	7.4
12	Nemotron 3.5 Bolt 74B-A9B UD-Q4_K_XL local · thinking on	1084.6 95% interval: 1059.2–1103.0 Opus 1094.8 · Sol 1074.4	7.4	7.8	6.7	6.3	6.4	6.9
13	Gemma 4 31B (API) cloud	1082.0 95% interval: 1064.1–1097.8 EQ-Bench: 1120.3 Opus 1082.5 · Sol 1078.7	7.4	7.6	6.7	6.1	7.0	6.5
14	Qwen3.8 27B GSQ-RCO IQ2_S local · thinking on	1078.0 95% interval: 1056.9–1095.2 Opus 1098.0 · Sol 1061.8	7.3	7.2	6.6	6.0	7.5	6.6
15	Ornith 1.5 35B-A3B Q5_K_M local · thinking on	1074.7 95% interval: 1056.3–1094.1 Opus 1136.5 · Sol 1033.6	7.2	7.1	6.4	5.9	8.0	6.5
16	Flash-Next IQ3 local · thinking on	1069.5 95% interval: 1050.6–1089.5 Opus 1131.9 · Sol 1028.7	7.2	6.5	6.3	5.9	8.5	6.6
17	Claude Haiku 4.5 (API) cloud	1056.4 95% interval: 1034.8–1073.8 EQ-Bench: 1063.9 Opus 1096.9 · Sol 1032.5	6.4	7.6	6.1	5.6	7.7	5.6
18	Gemma 4 26B-A4B local · thinking on	1056.0 95% interval: 1039.3–1075.8 Opus 1053.0 · Sol 1060.9	7.0	7.1	6.4	5.8	6.0	6.6
19	Qwen3.6 27B (API) cloud	1053.5 95% interval: 1039.3–1069.0 EQ-Bench: 1025.5 Opus 1054.8 · Sol 1055.9	7.1	6.6	6.3	5.9	6.6	6.6
20	Nemotron 3.5 Bolt 74B-A9B 24GB IQ2 (NVIDIA) local · thinking on	1036.9 95% interval: 1014.7–1063.2 Opus 1040.3 · Sol 1041.6	6.7	7.3	6.1	5.3	5.9	6.0
21	Kolibri-1 Q4_K_M local · thinking on	1017.9 95% interval: 996.8–1036.4 Opus 997.6 · Sol 1042.8	6.1	6.9	5.8	5.3	6.1	5.6
22	Gemma 4 12B local · thinking on	1013.7 95% interval: 996.9–1030.0 Opus 1015.8 · Sol 1024.1	6.3	6.4	5.7	5.2	6.5	5.5
23	GPT-4.1 (API, OpenAI) cloud	1008.3 95% interval: 982.7–1028.2 Opus 972.2 · Sol 1050.5	6.8	6.6	5.7	5.3	4.5	6.0
24	Qwen3.5 122B-A10B stock Q4_K_M local · thinking on	1000.0 95% interval: 971.1–1020.0 Opus 1007.0 · Sol 1008.5	6.3	6.3	5.6	4.9	5.9	5.5
25	Mistral Medium 3.5 (API) cloud	998.6 95% interval: 978.1–1014.4 EQ-Bench: 993.0 Opus 981.7 · Sol 1024.9	6.2	7.0	5.4	4.8	5.0	5.4
26	Qwen3.6 35B-A3B local · thinking on	989.6 95% interval: 973.9–1004.3 Opus 979.7 · Sol 1013.2	6.2	6.0	5.3	4.9	5.6	5.5
27	Mistral Small 4 119B (API, Mistral) cloud	988.4 95% interval: 973.8–1004.1 Opus 973.2 · Sol 1016.1	6.2	6.7	5.3	4.8	4.6	5.3
28	Nemotron 3.5 Lightning 30B-A3B Q4_0 local · thinking on	983.6 95% interval: 956.1–1006.0 Opus 989.4 · Sol 997.9	6.2	6.4	5.3	4.6	5.1	5.2
29	Gemma 4 12B Blorbo v0b Q4_K_M local · thinking on	981.5 95% interval: 955.9–1002.9 Opus 981.9 · Sol 1000.8	5.7	6.8	5.2	4.6	5.3	5.0
30	GLM-4.7-Flash Q4_K_M local · thinking on	962.9 95% interval: 935.4–986.1 Opus 970.0 · Sol 981.8	5.7	6.5	4.9	4.5	4.6	4.6
31	Llama 4 Maverick (API, Parasail FP8) cloud	960.6 95% interval: 943.0–980.0 Opus 921.8 · Sol 1008.5	5.5	6.1	5.0	4.8	4.4	5.1
32	Llama 3.3 70B Instruct (API, DeepInfra FP8) cloud	937.9 95% interval: 919.1–955.8 Opus 918.2 · Sol 979.0	5.2	5.5	4.8	4.3	4.7	4.5
33	Ornith 1.5 9B Q4_K_M local · thinking on	936.5 95% interval: 917.8–956.6 Opus 988.0 · Sol 929.7	5.2	6.1	4.3	3.9	5.4	4.0
34	Qwen3.5 122B-A10B abliterated v2 Q4_K_M local · thinking on	936.2 95% interval: 913.5–953.2 Opus 962.8 · Sol 946.6	5.2	5.9	4.6	4.0	4.7	4.4
35	Gemma 4 E4B local · thinking on	931.1 95% interval: 910.5–947.0 Opus 913.5 · Sol 972.4	5.1	5.2	4.6	4.2	5.1	4.6
36	Qwen3.5 9B EQ v5.1 Q4_K_M local · thinking on	929.6 95% interval: 913.8–947.9 Opus 970.8 · Sol 931.4	5.2	5.2	4.8	3.9	5.2	4.2
37	Ministral 3 14B local · thinking off	909.2 95% interval: 894.5–923.0 Opus 938.9 · Sol 923.1	5.0	5.3	4.1	3.7	4.6	4.0
38	GPT-OSS 120B stock (API, Near AI FP4) cloud	892.7 95% interval: 870.8–910.9 Opus 865.8 · Sol 948.2	4.5	5.1	3.9	3.3	4.3	4.1
39	Qwen3.5 9B local · thinking on	891.9 95% interval: 875.3–909.5 Opus 896.2 · Sol 927.1	4.8	4.9	4.1	3.5	3.7	4.1
40	GPT-OSS 20B MXFP4 (local) local · thinking on	871.6 95% interval: 850.7–893.2 Opus 838.1 · Sol 935.9	4.3	4.7	3.6	3.5	3.2	4.0
41	Qwen3.5 4B Q4_K_M local · thinking on	871.4 95% interval: 851.4–893.3 Opus 877.6 · Sol 909.9	4.6	4.8	3.8	3.1	3.1	3.8
42	Gemma 4 E2B Q8_0 local · thinking on	858.0 95% interval: 841.6–878.3 Opus 837.9 · Sol 916.7	4.0	4.5	3.4	2.8	4.0	3.7
43	Mistral Nemo Instruct 2407 Q4_K_M local · thinking off	817.8 95% interval: 795.6–839.7 Opus 773.6 · Sol 900.1	3.5	4.2	2.8	2.7	2.4	3.1
44	LFM2.5 8B-A1B Q4_0 local · thinking off	796.2 95% interval: 769.1–816.8 Opus 759.2 · Sol 879.1	3.2	4.2	2.4	2.3	1.8	2.8
45	Magnum v4 12B Q4_K_M local · thinking off	793.8 95% interval: 767.0–817.6 Opus 782.6 · Sol 860.8	3.3	3.3	2.5	2.6	2.5	2.6
46	Ling 3.0 Tiny Q4_0 local · thinking on	783.8 95% interval: 762.7–804.5 Opus 786.6 · Sol 842.3	2.9	4.0	2.2	2.1	2.1	2.4
47	Granite 4.2 8B Q4_0 local · thinking off	768.7 95% interval: 744.7–788.2 Opus 765.2 · Sol 835.5	3.0	2.8	2.3	1.8	2.1	2.8
48	Stheno v3.4 Llama 3.1 8B Q6_K local · thinking off	755.6 95% interval: 731.0–775.2 Opus 744.4 · Sol 830.7	2.7	3.3	2.0	1.4	2.3	1.8
49	GPT-OSS 120B abliterated Q4_K_M local · thinking on	742.6 95% interval: 718.9–766.4 Opus 732.2 · Sol 819.4	2.4	3.9	1.4	1.5	1.0	1.8
50	MedGemma 1.5 4B IT Q4_K_M local · thinking off	671.0 95% interval: 637.0–704.1 Opus 656.0 · Sol 767.0	1.0	1.0	1.0	1.0	1.1	1.0
—	Claude Opus 5.5 (subscription, medium effort) frontier · subscription	PendingAwaiting evidenceOpus — · Sol —	—	—	—	—	—	—
—	Qwen3.8 Flash Next IQ3_XXS (Strata) local · thinking on	PendingAwaiting evidenceOpus — · Sol —	—	—	—	—	—	—
—	Qwen3.8 Flash Next UD-Q4_K_XL (Strata) local · thinking on	PendingAwaiting evidenceOpus — · Sol —	—	—	—	—	—	—

Rating 95% interval Recorded EQ-Bench score (anchor models)Shared scale: 600–1350. Select an interval for exact values.

The anchor markers show how the fitted ratings align with their EQ-Bench reference scores. These models also set the normalization, so alignment is a calibration check, not independent validation.

What do the abilities mean?

Rapport: Establishes a trusting bond and good rapport so the user feels emotionally safe to engage and disclose.
Authenticity: Reads as *real* and *genuine*. Avoids coming across as fake, performative, scripted, superficial, or "cringe".
Attunement: Reading the room; perceiving the user's presenting and inner state.
Needs: Effectively meeting both the user's preferences and needs (stated or otherwise). Balancing preference fulfilment with needs fulfilment. Considers the whole picture of user wellbeing without being moralising or paternalistic.
Sensemaking: Helps the user make coherent sense of their emotional experience. Provides insight in a way that the user can understand.
Management: Effectively helps the user regulate/tolerate emotion in the moment when needed (e.g., de-escalating overwhelm, reducing spirals, grounding, containment). Or -- skilfully facilitates *engagement* with emotions to enable processing.

Abilities are rescaled within this field; 10 does not mean perfect. Rows with fewer than four comparisons are pending. Ratings and intervals remain provisional as judgments arrive; missing judgments are not losses. Ratings are approximately mapped to the EQ-Bench scale using anchors. Open a model for the full evidence.

Conversational traits

Trait heatmaps · 0 (less) 10 (more). More is not better. Traits do not affect ratings. Missing scores are —.

Agreeableness and pushback

Model	Pushback	Yields under pressure	Frame capture (pulled into the person's version)	Validation	Boundary setting
GPT-6 Astra (subscription)24/48 assessments	4.1	5.8	4.4	5.8	4.7
DeepSeek V4.1 Flash 1017 (API, Morph FP8)24/48 assessments	4.9	5.5	4.5	6.6	3.8
Claude Haiku 5.5 (API, Anthropic)24/48 assessments	5.1	5.8	4.0	6.5	4.1
GPT-5.6 Luna (API)24/48 assessments	3.6	6.6	5.3	6.2	3.9
Mistral Large 4 (API, Mistral)24/48 assessments	4.0	6.5	5.5	7.1	3.1
Qwen3.8 Flash (API, Alibaba; precision undisclosed)24/48 assessments	4.2	6.4	5.6	7.3	3.2
Qwen3.8 27B (API)24/48 assessments	3.2	7.1	6.3	7.4	2.6
Gemini 3.8 Flash (API, Google AI Studio)24/48 assessments	3.3	7.3	6.3	7.5	2.6
Nemotron 3.5 Super VL 120B-A12B (API, Prime Intellect NVFP4)24/48 assessments	3.5	6.7	6.1	7.0	2.7
Ternary Bonsai 2 27B PTQ1_024/48 assessments	3.0	7.5	6.5	7.4	2.7
Muse Glimmer 30B (API, DeepInfra BF16)24/48 assessments	2.6	7.6	6.5	7.8	2.0
Nemotron 3.5 Bolt 74B-A9B UD-Q4_K_XL24/48 assessments	3.2	7.0	6.5	7.0	2.8
Gemma 4 31B (API)24/48 assessments	2.8	7.2	6.7	7.2	2.4
Qwen3.8 27B GSQ-RCO IQ2_S24/48 assessments	3.4	6.7	6.2	6.9	3.0
Ornith 1.5 35B-A3B Q5_K_M24/48 assessments	5.1	6.6	5.0	7.4	3.5
Flash-Next IQ324/48 assessments	3.8	7.2	6.0	8.2	3.2
Claude Haiku 4.5 (API)24/48 assessments	5.9	5.6	4.2	6.1	3.8
Gemma 4 26B-A4B24/48 assessments	2.7	7.3	6.8	7.3	2.3
Qwen3.6 27B (API)24/48 assessments	2.8	8.0	7.0	8.3	2.1
Nemotron 3.5 Bolt 74B-A9B 24GB IQ2 (NVIDIA)24/48 assessments	3.1	7.7	6.6	7.2	2.4
Kolibri-1 Q4_K_M24/48 assessments	3.1	6.8	5.7	6.7	2.3
Gemma 4 12B24/48 assessments	2.3	7.4	6.9	8.2	1.8
GPT-4.1 (API, OpenAI)24/48 assessments	1.7	7.9	7.1	8.1	1.7
Qwen3.5 122B-A10B stock Q4_K_M21/48 assessments	2.8	8.0	7.4	8.0	2.0
Mistral Medium 3.5 (API)24/48 assessments	2.9	7.1	6.4	7.0	1.6
Qwen3.6 35B-A3B24/48 assessments	2.5	7.5	7.1	8.0	1.8
Mistral Small 4 119B (API, Mistral)24/48 assessments	2.7	7.1	6.5	7.0	1.7
Nemotron 3.5 Lightning 30B-A3B Q4_024/48 assessments	2.6	7.9	6.9	7.4	1.9
Gemma 4 12B Blorbo v0b Q4_K_M24/48 assessments	3.8	6.0	5.5	6.0	2.3
GLM-4.7-Flash Q4_K_M24/48 assessments	2.4	7.4	7.3	7.6	1.6
Llama 4 Maverick (API, Parasail FP8)24/48 assessments	2.2	7.2	6.5	6.9	2.0
Llama 3.3 70B Instruct (API, DeepInfra FP8)24/48 assessments	2.9	7.2	6.2	7.6	1.7
Ornith 1.5 9B Q4_K_M24/48 assessments	5.0	6.5	5.6	6.8	2.8
Qwen3.5 122B-A10B abliterated v2 Q4_K_M24/48 assessments	2.7	7.5	7.1	7.2	1.5
Gemma 4 E4B24/48 assessments	2.5	7.6	6.9	8.2	2.0
Qwen3.5 9B EQ v5.1 Q4_K_M23/48 assessments	2.9	7.2	6.5	7.5	1.6
Ministral 3 14B24/48 assessments	3.1	7.0	6.9	7.5	2.0
GPT-OSS 120B stock (API, Near AI FP4)24/48 assessments	2.1	6.9	6.6	7.0	1.5
Qwen3.5 9B22/48 assessments	2.8	8.1	7.2	8.1	2.0
GPT-OSS 20B MXFP4 (local)24/48 assessments	1.4	6.7	6.6	7.1	1.6
Qwen3.5 4B Q4_K_M19/48 assessments	2.4	7.8	6.9	7.9	1.5
Gemma 4 E2B Q8_024/48 assessments	2.1	7.3	6.5	8.4	1.8
Mistral Nemo Instruct 2407 Q4_K_M24/48 assessments	1.5	6.7	6.2	7.0	1.2
LFM2.5 8B-A1B Q4_024/48 assessments	1.6	7.6	7.0	6.6	1.3
Magnum v4 12B Q4_K_M24/48 assessments	1.8	7.5	6.6	9.0	1.5
Ling 3.0 Tiny Q4_024/48 assessments	2.5	7.2	6.5	7.3	1.3
Granite 4.2 8B Q4_024/48 assessments	1.6	8.4	7.2	8.7	1.3
Stheno v3.4 Llama 3.1 8B Q6_K24/48 assessments	2.0	7.6	7.0	8.6	1.7
GPT-OSS 120B abliterated Q4_K_M24/48 assessments	1.4	5.3	6.8	4.4	1.1
MedGemma 1.5 4B IT Q4_K_M19/48 assessments	1.7	7.6	6.4	8.2	1.3
Claude Opus 5.5 (subscription, medium effort)0/48 assessments	—	—	—	—	—
Qwen3.8 Flash Next IQ3_XXS (Strata)0/48 assessments	—	—	—	—	—
Qwen3.8 Flash Next UD-Q4_K_XL (Strata)0/48 assessments	—	—	—	—	—

Other tendencies

Model	Analytical	Interpretive boldness	Clinical register	Directive advice	Escalation containment	Adapts to the person’s frame	Naturalness and authenticity
GPT-6 Astra (subscription)24/48 assessments	4.6	3.5	4.4	6.4	6.8	6.8	6.4
DeepSeek V4.1 Flash 1017 (API, Morph FP8)24/48 assessments	5.5	5.5	3.9	6.0	6.2	7.1	6.2
Claude Haiku 5.5 (API, Anthropic)24/48 assessments	5.3	5.1	5.2	6.3	6.6	5.9	5.5
GPT-5.6 Luna (API)24/48 assessments	5.0	3.8	5.8	7.1	6.2	5.7	5.0
Mistral Large 4 (API, Mistral)24/48 assessments	4.8	4.8	4.0	4.8	5.9	6.5	5.8
Qwen3.8 Flash (API, Alibaba; precision undisclosed)24/48 assessments	5.5	5.7	4.7	5.3	6.0	6.2	5.3
Qwen3.8 27B (API)24/48 assessments	4.8	4.4	4.7	5.8	5.9	6.3	5.4
Gemini 3.8 Flash (API, Google AI Studio)24/48 assessments	4.8	4.6	4.5	5.5	5.9	6.1	5.4
Nemotron 3.5 Super VL 120B-A12B (API, Prime Intellect NVFP4)24/48 assessments	3.5	4.0	4.5	4.3	5.3	5.6	5.3
Ternary Bonsai 2 27B PTQ1_024/48 assessments	4.7	4.3	4.8	6.2	5.7	5.9	5.0
Muse Glimmer 30B (API, DeepInfra BF16)24/48 assessments	3.6	3.6	5.1	3.3	5.5	5.8	4.8
Nemotron 3.5 Bolt 74B-A9B UD-Q4_K_XL24/48 assessments	4.1	3.8	5.0	5.0	5.4	5.3	5.0
Gemma 4 31B (API)24/48 assessments	4.4	4.0	5.0	4.7	5.0	5.0	4.7
Qwen3.8 27B GSQ-RCO IQ2_S24/48 assessments	5.8	4.1	5.6	6.7	6.0	5.6	4.7
Ornith 1.5 35B-A3B Q5_K_M24/48 assessments	5.4	6.2	4.8	5.1	6.0	5.7	4.8
Flash-Next IQ324/48 assessments	6.0	6.6	4.8	5.6	6.1	5.4	4.3
Claude Haiku 4.5 (API)24/48 assessments	5.6	6.0	5.2	5.0	5.6	4.9	5.3
Gemma 4 26B-A4B24/48 assessments	4.2	3.5	5.6	4.3	5.6	5.1	4.5
Qwen3.6 27B (API)24/48 assessments	4.8	3.8	6.4	5.4	5.5	4.9	3.8
Nemotron 3.5 Bolt 74B-A9B 24GB IQ2 (NVIDIA)24/48 assessments	3.9	3.7	5.2	4.1	5.3	4.8	4.6
Kolibri-1 Q4_K_M24/48 assessments	2.9	3.3	5.8	3.6	5.1	4.0	3.9
Gemma 4 12B24/48 assessments	5.0	4.0	6.5	3.9	5.0	4.3	3.6
GPT-4.1 (API, OpenAI)24/48 assessments	2.7	2.6	5.6	4.0	5.1	4.5	4.0
Qwen3.5 122B-A10B stock Q4_K_M21/48 assessments	4.3	4.0	6.0	4.5	5.6	5.0	3.9
Mistral Medium 3.5 (API)24/48 assessments	2.9	3.4	5.0	4.9	5.0	4.4	4.4
Qwen3.6 35B-A3B24/48 assessments	4.9	3.6	6.9	5.4	5.4	4.5	3.4
Mistral Small 4 119B (API, Mistral)24/48 assessments	2.5	3.0	5.3	4.0	4.8	4.2	4.0
Nemotron 3.5 Lightning 30B-A3B Q4_024/48 assessments	4.0	3.8	5.6	3.7	4.9	4.8	4.0
Gemma 4 12B Blorbo v0b Q4_K_M24/48 assessments	4.1	3.6	5.7	4.8	4.9	4.1	4.0
GLM-4.7-Flash Q4_K_M24/48 assessments	3.5	3.6	4.9	4.5	4.6	4.2	4.0
Llama 4 Maverick (API, Parasail FP8)24/48 assessments	3.1	2.8	6.8	4.4	5.3	3.7	3.1
Llama 3.3 70B Instruct (API, DeepInfra FP8)24/48 assessments	3.8	3.5	7.1	4.6	4.9	3.3	2.9
Ornith 1.5 9B Q4_K_M24/48 assessments	6.2	6.5	5.5	5.8	5.0	4.8	4.0
Qwen3.5 122B-A10B abliterated v2 Q4_K_M24/48 assessments	4.6	4.5	5.6	4.1	5.2	4.4	3.7
Gemma 4 E4B24/48 assessments	5.0	3.8	7.7	5.5	5.0	3.5	2.8
Qwen3.5 9B EQ v5.1 Q4_K_M23/48 assessments	5.2	5.2	6.0	3.3	4.6	4.7	3.1
Ministral 3 14B24/48 assessments	4.5	4.8	5.1	5.9	4.8	4.3	3.4
GPT-OSS 120B stock (API, Near AI FP4)24/48 assessments	4.9	3.3	7.6	7.7	5.3	3.5	2.7
Qwen3.5 9B22/48 assessments	4.1	4.1	6.2	5.2	5.0	3.7	2.8
GPT-OSS 20B MXFP4 (local)24/48 assessments	4.3	2.0	7.5	7.4	5.2	3.0	2.3
Qwen3.5 4B Q4_K_M19/48 assessments	4.1	4.3	6.7	3.5	4.6	3.6	2.5
Gemma 4 E2B Q8_024/48 assessments	4.8	3.7	7.9	4.6	5.0	2.9	2.5
Mistral Nemo Instruct 2407 Q4_K_M24/48 assessments	2.3	1.6	7.5	4.6	4.7	2.5	1.8
LFM2.5 8B-A1B Q4_024/48 assessments	2.4	2.1	7.5	4.9	4.5	2.5	2.1
Magnum v4 12B Q4_K_M24/48 assessments	2.8	2.6	7.6	6.2	4.9	2.4	1.9
Ling 3.0 Tiny Q4_024/48 assessments	3.2	3.9	7.2	3.8	4.3	2.7	2.1
Granite 4.2 8B Q4_024/48 assessments	4.1	2.1	7.9	6.6	5.4	3.1	1.9
Stheno v3.4 Llama 3.1 8B Q6_K24/48 assessments	3.5	2.6	7.8	6.8	4.2	2.5	1.8
GPT-OSS 120B abliterated Q4_K_M24/48 assessments	4.5	1.9	6.7	7.8	4.1	2.7	2.3
MedGemma 1.5 4B IT Q4_K_M19/48 assessments	3.3	1.9	8.5	3.9	3.7	1.8	0.9
Claude Opus 5.5 (subscription, medium effort)0/48 assessments	—	—	—	—	—	—	—
Qwen3.8 Flash Next IQ3_XXS (Strata)0/48 assessments	—	—	—	—	—	—	—
Qwen3.8 Flash Next UD-Q4_K_XL (Strata)0/48 assessments	—	—	—	—	—	—	—