Grok 5 Mini

What it is

Grok 5 Mini is xAI's small tier — the model the note on this page positions for routing and lightweight agents. It shares the Grok 5 line's general shape at a fraction of the capability ceiling, and the trade is visible in the numbers: everything sits in the high seventies except Speed at 91, which is the reason the model exists.

As with the rest of xAI's line, weights and training data are not published and this board has no price or index for it, so the composite is the only hard measurement available here.

How the composite is built

Overall 80 weights Coding 24%, Terminal 20%, Reasoning 20%, Tool use 15%, Context 13%, Speed 8%. The five capability axes cluster between 77 and 82 — a flat profile that reads as reliable-at-a-middling-level rather than good at anything in particular.

Coding 79

Level with Phi-5 and Mistral Small 3.2, and eleven behind Grok 5's 90. Boilerplate and glue code, not architecture work.

Terminal 77

Four below Claude Haiku 4.5's 81, the nearest small-model comparison, and eleven behind Grok 5. Keep it on short, well-specified shell tasks.

Reasoning 79

Five behind Phi-5's 84, which is the small tier's reasoning outlier here. Adequate for classification and triage; thin for multi-step planning.

Tool use 78

Seven behind Haiku 4.5's 85 and level with Gemma 3 27B's 78. Expect more malformed calls in long chains than the 85-plus models produce.

Context 82

Level with Llama 5 Scout and Gemma 3 27B, and seven behind GPT-5.6 Luna's 89 — unusual, since Luna is the comparable cost tier.

Speed 91

The point of the model: behind only Haiku 4.5's 94 anywhere on this board, and level with the fastest cost tiers. This is what high-volume routing buys.

Where it fits

Routing, classification, extraction and the many small calls inside a bigger system — the same job Haiku 4.5 and Luna do. It is a reasonable default inside an xAI-centred stack where you want one vendor's billing and one vendor's tool schema, with Grok 5 handling anything the Mini cannot.

Limits

• No price on this board. With no rate and no Artificial Analysis index, the Value column is Unrated and cost comparisons against Luna or Haiku 4.5 have to be made on the vendor's own site.
• Nothing here is class-leading except Speed. Coding 79, Reasoning 79 and Tool use 78 are all mid-tier; if the router hands it hard work, you will see it.
• Closed weights. No self-hosting option and no exit route except a different provider's API.
• Context 82 is beaten by the cost-tier competition — Luna's 89 in particular. Long-document routing is not this model's strength.

Price and access

No published price, no index, and no gateway listing on this board — access is through xAI directly. Several later Grok listings do appear on OpenRouter, OpenCode Zen and OpenCode Go, so third-party availability is worth checking at the point of purchase rather than assumed. Last scored 15 Sep 2026.

Alternatives on this board

• Claude Haiku 4.5 — 80 at $1/$5 with Speed 94 and Tool use 85. The strongest small tier at tool work, priced here.
• GPT-5.6 Luna — 81 at $0.20/$1.20 with Context 89. Better on paper and published value at a known rate.
• Mistral Small 3.2 — 79 at $0.09/$0.25 with open weights, if the routing tier must also be self-hostable.
• Grok 5 — 87, the escalation path inside the same vendor.

Sources