TMB Scoreboard — Planning & Agents

The Monocle Bear · updated 29 July 2026

• Axes: plan_decompo, plan_spec, plan_judgment, reasoning, calc, agent_exec, agent_safety (7 axes)
• Evaluator: Claude Opus 4.8 (rejudge pass)
• Scale: 0–100 per axis

---

Planning & Agent scoreboard

#	Model	Avg	P.decompo	P.spec	P.judgment	Reasoning	Calc	Agent exec	Agent safe	TPS
1	Kimi-K3-OR	97.7	100	94	100	94	98	98	100	31.2
2	Qwen3.8-prev ALI 🏠	97.3	98	96	99	100	98	96	94	44.8
3	o3-OR	96.3	98	100	95	90	96	98	98	132.8
4	Nex N2 pro OR	96.2	96	98	88	96	96	100	99	134.1
5	nemotron3 super OR	95.8	96	94	89	100	100	92	99	112.5
6	RING2.6 OR	95.3	92	94	94	94	98	96	98	125.5
7	kimi-k2.6-OR	95.3	84	98	92	98	100	100	95	42.7
8	deepseek v4 pro OR	95.1	88	98	86	100	100	95	99	52.2
9	qwen3.5 397 OR	94.7	84	94	89	100	100	96	100	50.9
10	glm-5.1-OR	94.6	82	96	88	100	100	96	100	52.4
11	KIMI 2.7 OR	94.2	100	94	82	94	96	96	98	48.4
12	Ling-3-flash - OR	94.0	92	96	87	94	99	91	99	298.9
13	step3.7flash OR	94.0	96	—	85	94	98	91	100	170.9
14	Qwen3.7+ OR	93.7	78	92	96	94	100	98	98	54.1
15	GLM5.2 Q6 🏠	93.1	86	96	78	100	98	94	100	11.2
16	Qwen3.7Max OR	92.3	78	94	84	94	100	98	99	54.0
17	GLM5.2-OR	91.9	78	92	78	100	100	96	99	41.9
18	minimax2.7 MI	91.9	90	94	97	94	94	78	96	48.4
19	aion3 OR	89.7	86	82	76	100	96	91	97	46.4
20	mimo2.5 XI	89.4	86	98	94	54	100	95	99	75.0
21	kat-coder-pro-v2.5-OR	88.2	80	98	87	56	100	99	98	109.1
22	minimax3 MI	88.0	94	96	93	46	98	90	99	68.1
23	Hy3 - OR	87.6	86	96	79	64	96	92	100	51.4
24	Ling-2.6 - OR	86.4	98	98	87	30	99	95	98	108.3
25	mercury-2-OR	85.4	88	94	81	48	96	99	92	433.9
26	Qwen3-Next-80B-A3B-Instruct-8bit 🏠	85.1	78	96	92	34	99	96	100	61.3
27	Qwen3-coder-Next-80b 🏠	82.4	76	100	81	36	98	89	97	58.6
28	mistral L3 OR	82.3	84	80	83	40	96	93	100	64.8
29	Qwen3.5-122B-8H16 🏠	82.2	94	96	77	40	100	90	79	41.2
30	destral2 OR	81.5	80	96	76	34	92	98	95	71.3
31	laguna-xs-2.1 🏠	80.9	78	96	80	41	92	95	84	79.4
32	deepseek v4 flash OR	80.2	88	68	76	34	100	96	99	58.2
33	lagunaM1 OR	79.3	72	94	78	22	92	99	98	42.2
34	mistral M3.5	79.2	78	80	86	32	93	89	97	144.9
35	Kimi-Linear-48B 🏠	76.9	78	88	69	26	98	93	86	82.1
36	Laguna-S2.1 🏠	73.4	66	96	78	18	76	89	91	42.4

---

References

Via OpenRouter — directly comparable with contenders.

Model	Avg	P.decompo	P.spec	P.judgment	Reasoning	Calc	Agent exec	Agent safe
Fable5-OR-REF	98.5	100	100	100	100	100	98	92
Opus-4.8-OR-REF	98.1	98	100	98	100	98	96	96
GPT 5.6 Sol-REF	95.7	100	98	96	82	100	96	98
GPT 5.5-REF	94.5	100	98	87	82	100	96	99
Opus5-OR-REF	94.4	94	88	100	88	97	95	99
Fusion-REF	93.6	94	88	100	94	98	97	84

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	P.decompo	P.spec	P.judgment	Reasoning	Calc	Agent exec	Agent safe
Opus4.8-REF	97.5	94	96	100	94	100	100	98
Fable5-REF	97.5	100	98	95	94	100	96	99

---

TMB Benchmark v5 — The Monocle Bear — July 2026.