Grok 5

What it is

Grok 5 is xAI's frontier tier and the highest-scoring non-Anthropic, non-Google model on this board at overall 87. The note on this page describes it as fast for its tier with strong reasoning and current-events grounding — the latter being the Grok line's long-standing position, since the models are trained inside a company built around a live social feed.

xAI publishes little about parameters or training data, so treat architecture claims as unverified. What the board measures is the behaviour of the hosted model: six axes, one composite, no vendor numbers mixed in.

How the composite is built

Overall 87 weights Coding 24%, Terminal 20%, Reasoning 20%, Tool use 15%, Context 13%, Speed 8%. Grok 5 is even across five axes from 87 to 91, with Speed at 70 the single point that pulls the composite down.

Coding 90

Level with Sonnet 5 and Qwen 4 Coder, three behind Sol's 93 and six behind Opus 5's 96. Genuine frontier-class code work.

Terminal 88

Level with Nemo 3 Ultra and behind only the top terminal tiers, from Sol's 89 to Opus 5's 95. One of the stronger autonomous-operation scores outside Anthropic.

Reasoning 91

Its best axis and the strongest anywhere here outside the 93-plus group. Two behind Gemini 3.6 Pro's 93, two ahead of GPT-5.6 Terra's 89.

Tool use 87

One ahead of Terra's 86 and four behind Sol's 91. Solid orchestration, though the board's top tool scores all sit in the 90s.

Context 88

Mid-frontier: two behind Terra's 90 and ten behind Gemini 3.6 Pro's 98. Fine for repository-level reads, not whole-codebase reasoning.

Speed 70

Quicker than the other frontier tiers — Opus 5's 62, Sol's 60, the 48s — but ten behind Terra and twenty behind the small tiers. Frontier output still costs latency.

Where it fits

General reasoning and coding work where you want frontier quality without the Anthropic or Google stack, and anything where current-events grounding genuinely matters. It is also the natural centre of gravity if your organisation is already buying xAI inference for other products.

Limits

• No price is published on this board — the Value column reads Unrated because there is no rate or Artificial Analysis index to divide by, not as a judgement on cost.
• Closed weights, closed data, and no self-hosting route. Everything depends on xAI's endpoint and its uptime.
• Speed 70 rules out the cheapest high-volume patterns; route those to Grok 5 Mini at 91.
• The current-events claim is marketing until you test it. Live-feed grounding is also a bias source — verify on your own queries before relying on it.

Price and access

No price or index is listed for Grok 5 on this board, so no blended rate can be computed — check xAI's own site for current numbers. This board shows no gateway listing for Grok 5 itself, though later Grok listings appear on OpenRouter, OpenCode Zen and OpenCode Go. xAI iterates quickly and several Grok listings sit unscored alongside this one. Last scored 15 Sep 2026.

Alternatives on this board

• GPT-5.6 Sol — 90 at $2/$10, the priced frontier tier against which Grok 5's value can be judged.
• Gemini 3.6 Pro — 89 with Context 98 and Reasoning 93, the other non-Anthropic frontier option.
• Nemo 3 Ultra — 86 at $0.60/$2.40 with Terminal 88, if agentic pipelines are the workload and price matters.
• Grok 5 Mini — 80 with Speed 91, the same vendor's answer for volume.

Sources