TMB Scoreboard — Coding

The Monocle Bear · updated 29 July 2026

• Axes: python, code_general, debug, react, swift, refactoring (6 axes, /100 each)
• Evaluator: Claude Opus 4.8 (rejudge pass)
• Scale: 0–100 per axis

---

Code scoreboard

#	Model	Avg code	Python	Code gen.	Debug	React	Swift	Refact.	TPS
1	Kimi-K3-OR	98.0	98	98	98	98	99	97	31.2
2	Qwen3.8-prev ALI 🏠	97.5	97	98	100	97	96	97	44.8
3	glm-5.1-OR	95.4	96	94	100	91	93	98	52.4
4	kimi-k2.6-OR	94.5	97	95	93	94	92	96	42.7
5	Nex N2 pro OR	94.3	96	94	96	98	88	—	134.1
6	deepseek v4 pro OR	93.9	95	91	100	98	84	96	52.2
7	GLM5.2-OR	93.9	97	92	98	94	88	95	41.9
8	nemotron3 super OR	92.6	96	92	96	90	88	94	112.5
9	KIMI 2.7 OR	92.5	92	92	98	90	90	93	48.4
10	Qwen3.7+ OR	92.2	95	91	92	98	85	92	54.1
11	aion3 OR	91.9	93	91	100	93	86	88	46.4
12	Qwen3.7Max OR	90.3	95	89	88	92	93	85	54.0
13	deepseek v4 flash OR	88.2	94	88	86	87	82	92	58.2
14	minimax3 MI	87.9	95	86	94	94	58	100	68.1
15	Ling-3-flash - OR	85.8	94	83	100	90	62	86	298.9
16	mimo2.5 XI	85.7	96	86	93	96	45	98	75.0
17	minimax2.7 MI	84.9	96	86	81	90	85	72	48.4
18	GLM5.2 Q6 🏠	84.0	90	82	86	86	75	85	11.2
19	mercury-2-OR	83.9	84	82	94	74	69	100	433.9
20	step3.7flash OR	83.5	88	84	82	86	80	82	170.9
21	qwen3.5 397 OR	83.2	75	81	100	90	74	80	50.9
22	Ling-2.6 - OR	81.6	94	83	89	84	76	64	108.3
23	laguna-xs-2.1 🏠	81.4	93	82	78	88	67	80	79.4
24	Hy3 - OR	81.1	95	81	78	80	65	87	51.4
25	Qwen3-coder-Next-80b 🏠	79.6	88	81	77	89	62	80	58.6
26	mistral M3.5	79.4	83	78	72	88	65	90	144.9
27	mistral L3 OR	77.1	93	76	84	86	71	52	64.8
28	o3-OR	76.7	94	76	92	88	35	76	132.8
29	Qwen3-Next-80B-A3B-Instruct-8bit 🏠	76.5	91	78	80	82	68	60	61.3
30	Qwen3.5-122B-8H16 🏠	75.4	94	76	74	78	60	70	41.2
31	Laguna-S2.1 🏠	74.5	83	71	84	75	50	83	42.4
32	RING2.6 OR	73.1	89	77	88	32	70	83	125.5
33	lagunaM1 OR	69.7	80	70	81	72	43	72	42.2
34	Kimi-Linear-48B 🏠	65.2	80	68	64	41	63	75	82.1
35	destral2 OR	65.2	86	68	56	78	54	50	71.3
36	kat-coder-pro-v2.5-OR	63.0	92	55	91	24	36	80	109.1

---

References

Via OpenRouter — directly comparable with contenders.

Model	Avg code	Python	Code gen.	Debug	React	Swift	Refact.
Fusion-REF	97.7	99	98	100	96	96	98
GPT 5.6 Sol-REF	97.4	97	97	98	98	98	96
Fable5-OR-REF	96.7	97	97	98	97	96	96
GPT 5.5-REF	96.6	97	96	98	98	95	96
Opus5-OR-REF	93.0	98	93	99	96	72	99
Opus-4.8-OR-REF	92.8	97	93	92	93	88	93

Run under Claude Code harness (not directly comparable)

These runs went through the Claude Code harness rather than direct API — not on the same footing as the contenders above.

Model	Avg code	Python	Code gen.	Debug	React	Swift	Refact.
Fable5-REF	98.2	99	98	100	96	97	99
Opus4.8-REF	97.4	97	97	100	98	96	96

---

TMB Benchmark v5 — The Monocle Bear — July 2026.