terminal bench Benchmark: Scores and Sources

Published result; benchmark version and evaluation conditions remain in the id and result note.

pillar: coding · weight 1 within pillar · unit: % (higher is better) · official board ↗

Row	Model	Result	Normalized (0–100)	Confidence
1	Fugu Ultra sakana	82.1%	82.1	100% confidence 100 percent, Full
2	Fugu sakana	80.2%	80.2	100% confidence 100 percent, Full
3	Ornith 1.0 397B deepreinforce	78.2%	78.2	100% confidence 100 percent, Full
4	Ornith 1.0 397B deepreinforce	77.5%	77.5	100% confidence 100 percent, Full
5	Ornith 1.0 35B deepreinforce	64.2%	64.2	100% confidence 100 percent, Full
6	Ornith 1.0 35B deepreinforce	62.8%	62.8	100% confidence 100 percent, Full
7	Ornith 1.0 9B deepreinforce	43.1%	43.1	100% confidence 100 percent, Full
8	Ornith 1.0 9B deepreinforce	40.6%	40.6	100% confidence 100 percent, Full
9	GPT-5-Codex OpenAI	37.9%	37.9	6% confidence 6 percent, Low
10	Step 3.7 Flash stepfun	35.6%	35.6	13% confidence 13 percent, Low
11	GLM-4.7 Z.ai	33.4%	33.4	69% confidence 69 percent, Medium
12	Step 3.5 Flash 2603 stepfun	32.6%	32.6	13% confidence 13 percent, Low
13	Step 3.5 Flash stepfun	27.3%	27.3	88% confidence 88 percent, High
14	Gemini 2.5 Pro Google	26.5%	26.5	90% confidence 90 percent, High
15	GLM-4.6 Z.ai	25%	25.0	55% confidence 55 percent, Medium
16	GLM-4.5 Z.ai	22%	22.0	47% confidence 47 percent, Low
17	Qwen3 Max Alibaba / Qwen	20.5%	20.5	69% confidence 69 percent, Medium
18	GLM-4.5-Air Z.ai	20.5%	20.5	37% confidence 37 percent, Low
19	Devstral 2 mistral	18.9%	18.9	19% confidence 19 percent, Low
20	Mistral Small 4 mistral	17.4%	17.4	6% confidence 6 percent, Low
21	Mistral Large 3 mistral	15.9%	15.9	37% confidence 37 percent, Low
22	Qwen3-Coder 30B-A3B Instruct Alibaba / Qwen	15.2%	15.2	19% confidence 19 percent, Low
23	Gemini 2.5 Flash Google	13.6%	13.6	47% confidence 47 percent, Low
24	GPT-4o (2024-08-06) OpenAI	8.3%	8.3	100% confidence 100 percent, Full
25	GPT-4o (2024-11-20) OpenAI	8.3%	8.3	61% confidence 61 percent, Medium
26	DeepSeek-R1 DeepSeek	6.1%	6.1	88% confidence 88 percent, High
27	Mistral Large 2.1 mistral	6.1%	6.1	100% confidence 100 percent, Full
28	GLM-4.5V Z.ai	5.3%	5.3	37% confidence 37 percent, Low
29	Gemini 2.5 Flash-Lite Google	4.5%	4.5	6% confidence 6 percent, Low
30	Mistral Medium 3 mistral	3.8%	3.8	85% confidence 85 percent, High
31	Llama-3.3-70B-Instruct Meta	3%	3.0	100% confidence 100 percent, Full

Results are as published by the source behind each value (hover or tap the number). Benchmark scores are shown individually for every source/variant (a model may have multiple rows), and vary by version, harness and date; the normalized column uses the method’s fixed 0–100 scales and feeds the coding pillar of the SI Score.