AI Models Tier List for OpenClaw Users

OpenClaw's top 20 models ranked S to D — Sonnet 4.6, Gemini 3 Flash Preview, and Opus 4.6 take the top spots while Trinity Large Preview and GLM flagships land in D.

The S-Tier Pick: Claude Sonnet 4.6 Over Opus 4.6

The creator runs every OpenClaw orchestrator agent on Opus 4.6 or Sonnet 4.6 and calls Sonnet 4.6 the more practical choice. He states Sonnet 4.6 handles "80 to 90% of what Opus can do, but at a fraction of a cost," and recommends Opus 4.6 specifically to beginners for problem-solving, not as a default for long-running workflows.

Free Models That Punch Above Their Weight

• Gemini 3 Flash Preview is the other S-tier entry, praised for being both "broadly useful and legitimately high-end."
• Step 3.5 Flash is the most-used model on OpenClaw at $0 per million input and output tokens. It earns A-tier for being "fast, cheap, and surprisingly capable for agent workflows," though it "starts to wobble" on complex system-prompt-heavy work.
• Hunter Alpha enters A-tier with a 1M context window at zero cost, but the creator warns you "must be familiar with frontloading and backloading your critical information in those sessions."
• Kimi K2.5 also lands in A-tier for "punching above its cost bracket" and adds image-watching capability.

The D-Tier Traps

Trinity Large Preview is free but capped at 131K context. The creator argues the "dumb zone" hits fast versus the 200K benchmark he uses for Claude and MiniMax. GLM 4.5 Air and GLM 5 Turbo also land in D-tier — GLM 5 flagships cost $0.72/$2.30 per million tokens, and community feedback is "mostly meh."

MiniMax M2.5: The Creator's Personal S-Tier

The creator personally rates MiniMax M2.5 as S-tier because it tops OpenRouter's most-popular filter and "is like the king" for task execution by experienced developers, but community polling keeps it at B-tier — beginners reportedly struggle with its architecture handling.

Models Stuck In The Middle

GPT-5 Nano and Claude Sonnet 4.5 take A-tier as reliable specialists. Sonnet 4.5 is what the creator uses personally for Underlord AI video editing "because of the cost." Deepseek V3.2 and Gemini 3.1 Pro Preview sit at B-tier. C-tier holds Gemini 2.5 Flash, Claude Haiku 4.5, Grok 4.1, and GLM 5 — Gemini 2.5 Flash alone consumed 188B tokens this month despite being considered overtaken.