Agentic coding planner
The cheapest model good enough for agentic coding.
Coding capability × cost × tool-calling × whether it runs on your GPU — the join no other tool makes. Agents burn tokens, so $/M and decode speed matter more than for chat.
0.40
0 tok/s
Requires a GPU it fits on.
Capability vs cost · top-left is cheaper & better
13 models with a run cost150 models match = best value (coding per $)
| # | Model | Caps | |||||
|---|---|---|---|---|---|---|---|
| 1 | deepseek-ai/deepseek-v3.1 | 0.72 | $0.33 | 2.2 | |||
| 2 | stepfun-ai/step3 | 0.63 | $0.30 | 2.1 | |||
| 3 | deepseek-ai/deepseek-v3.1-terminus | 0.72 | $0.36 | 2.0 | |||
| 4 | arcee-ai/trinity-large-thinking | 0.65 | $0.39 | 1.7 | |||
| 5 | nex-agi/nex-n2-pro | 0.65 | $0.44 | 1.5 | |||
| 6 | minimaxai/minimax-text-01 | 0.45 | $0.43 | 1.1 | |||
| 7 | inclusionai/ring-2.6-1t | 0.74 | API | $0.85 | 0.9 | ||
| 8 | deepseek-ai/deepseek-v3.2-speciale | 0.74 | $0.85 | 0.9 | |||
| 9 | deepseek-ai/deepseek-prover-v2-671b | 0.70 | $0.92 | 0.8 | |||
| 10 | inclusionai/ring-1t | 0.74 | API | $0.98 | 0.8 | ||
| 11 | moonshotai/kimi-k2-thinking | 0.60 | $0.86 | 0.7 | |||
| 12 | meta-llama/llama-3.1-405b-fp8 | 0.55 | $1.95 | 0.3 | |||
| 13 | coherelabs/command-a-plus-05-2026-bf16 | 0.59 | $4.38 | 0.1 | |||
| 14 | willfalco/glm-5.2-exl3-tr3-3.36bpw | 0.73 | — | — | |||
| 15 | 0xsero/deepseek-v4-flash-0731-reap | 0.63 | — | — |
1–15 of 150 models
1 / 10
Coding scores, tool-calling flags, and prices are live data. Local $/M and speed = estimated decode at Q4_K_M / 8K on the selected GPU. Value = coding score per cheapest $/M (API or local, whichever is lower — the cheaper cell is highlighted).