Inkling 2 Small vs Open-Weight Rivals: Ripple-65B, Kimi K3.1 Pro, Hy4 and DeepSeek — How to Choose

"Strongest open model outside China" is a qualified title — drop the qualifier, and Inkling 2 Small's place in the open-weights field gets more complicated. This article puts five contemporaries on the same table — DeepSeek V4.1 Flash, Kimi K3.1 Pro, Tencent Hy4, GLM-5.3-Flash and Ripple-65B Preview — across five dimensions: AA index, Terminal-Bench, price, license and hardware bar. The headline: there is no all-around winner; Inkling 2 Small's relative edge is vision, finance and ARC reasoning, plus the Apache 2.0 + downloadable + free-entry combination — but on terminal-agent work it trails nearly every rival.

The table: five dimensions

"—" means the item was not confirmed in this round of checks, not that it does not exist. Ripple-65B's 80.52 is on Terminal-Bench 2.1, a different test suite from 4.0 — not directly comparable.

vs DeepSeek V4.1 Flash: the closest rival, ahead almost everywhere

AA index 32.6 vs 31, Terminal-Bench 4.0 33.8% vs 13.1%, FrontierCode 28.5% vs 25.4%, DeepSWE 77.0% vs 72.1% — DeepSeek wins narrowly across the coding tests; only Vals Finance (60.9% vs 61.5%) and vision tests favor Inkling. The two occupy nearly the same niche (light, efficient open MoEs), so the choice is task-driven: terminal agents and pure coding favor DeepSeek; vision and finance favor Inkling.

vs Kimi K3.1 Pro: the terminal-task gap

Kimi K3.1 Pro's Terminal-Bench 4.0 is 65.7%, five times Inkling 2 Small's 13.1%; its GPQA-Diamond 94.2% is also top-tier. On Plotly's data-analysis benchmark Kimi scored 42/43 (Inkling 39/43), but the full run cost $13.47 (Inkling $1.81) at about 2.5 minutes per question (Inkling 13 seconds). Kimi is a much bigger model — 2.8T total / 104B active — more capable, with a bigger bill and slower clock. Heavy terminal-agent work favors Kimi; speed and cost favor Inkling.

vs Tencent Hy4: the rival whose weights have not arrived

Hy4 scores 44.4% on Terminal-Bench 4.0 in Tencent's official appendix (Artificial Analysis independently measured 37% under different conditions), a 770B/49B MoE at $0.83/$2.50, with vision and computer-use skills. But its weights are only promised "shortly" — today it is Tencent Cloud hosting only. Inkling's edge is weights already in hand under Apache 2.0 for commercial use; Hy4's edge is the terminal-task score. Note Hy4 is not the cheaper option: its $0.83 input and $2.50 output prices are both above Inkling's $0.30/$1.00.

vs GLM-5.3-Flash: ahead on AA, behind on terminal work

AA index 31 vs 27.8 favors Inkling; Terminal-Bench 4.0 at 13.1% vs 32.8% favors GLM by a wide margin. On EnterpriseBench-CoreCraft, Inkling leads 71.8% to 46.7% — a clear edge on enterprise document work. Both are light open-weights models; choose by task: enterprise documents and vision favor Inkling, terminal agents favor GLM.

vs Ripple-65B Preview: the 65B dense challenger

Ripple-65B is a 65B dense open model released November 4 (Apache 2.0), scoring 80.52 on Terminal-Bench 2.1 and claiming to beat 500B–975B MoEs; Together hosts it at $0.20/$0.60, cheaper than Inkling. But three caveats: first, 80.52 is Terminal-Bench 2.1, not directly comparable with Inkling's 4.0 score; second, Ripple concedes it trails on SWE-bench Verified, DeepSWE and other tests; third, on Plotly's data-analysis benchmark, Ripple is the only U.S. open model besides Inkling to get all five statistics traps right — the two are in the same tier for data analysis. A 65B dense model's deployment bar is also far lower than a 594B MoE's, which is Ripple's practical edge.

The distribution dimension: free entry points and OpenCode

Beyond scores, access channels matter for selection. Inkling 2 Small has a free variant on OpenCode (inkling-2-small-free, limited time) and another on OpenRouter; Google Trends shows opencode among Inkling 2 Small's top related queries (3rd over 7 days), suggesting many users first meet the model through OpenCode. Weights already published + several hosts at the same price + free entry points is a fairly complete package among current open models — Hy4's weights have not arrived, and Kimi's and DeepSeek's free entry points differ, so "can I try it free" is worth factoring into the choice.