The arena · AI model review

Grok 4.3

by xAI

xAI's 1M-context frontier model at $1.25/$2.50 per 1M tokens, tuned for agentic tool use.

Arena score 4/5
Reasoning4.0
Coding3.5
Writing4.5
Speed3.5
Value4.5
Visit Grok 4.3agentic tool-calling pipelineshigh-volume cost-sensitive API workloadslong-context analysisconversational assistants
Price

$2.50/1M out

Flat $1.25 in / $2.50 out across all reasoning-effort tiers; cached input $0.20/1M. xAI's own pricing page shows no long-context surcharge, but OpenRouter lists tiered higher rates above 200K total tokens.

Provider

xAI

Context window

1M tokens

Input price

$1.25/1M in

Output price

$2.50/1M out

Modalities

text, vision (text output only)

Open weights

No

What is Grok 4.3?

xAI's flagship chat model (beta April 17, 2026), succeeding Grok 4.20. 1M-token context, text and image input, configurable reasoning effort (none/low/medium/high) in a single model. Positioned as a cheap near-frontier option for agentic and high-volume API workloads.

Grok 4.3 pros & cons

Pros

  • Aggressive pricing: $1.25/$2.50 per 1M tokens, 58% cheaper input and 83% cheaper output than Grok 4, undercutting GPT-5.5 and Gemini 3.1 Pro
  • Major agentic leap: +321 Elo on GDPval-AA versus Grok 4.20, with strong tool calling and instruction following
  • Cached input at $0.20/1M (84% discount), a big saver for repeated agent loops
  • Configurable reasoning effort (none/low/medium/high) in one model at one price, no routing between fast and deep variants
  • Praised on HN for natural, concise tone and token-dense outputs that lower real-world costs
  • Solid throughput around 130 output tokens/sec (Artificial Analysis)

Cons

  • Coding reasoning judged 'not competitive with the big April releases' by HN developers; intelligence frontier barely moved since Grok 4
  • Non-hallucination score dropped 8 points vs Grok 4.20 on AA-Omniscience; 4.20 remains xAI's safer pick for precision-critical domains
  • High time to first token (~13s at high reasoning effort per Artificial Analysis), painful for interactive apps
  • Context window shrank to 1M from Grok 4.20's 2M
  • Recurring trust and safety complaints (harmful content reports, inconsistent behavior) and no MCP/connected-apps support in the consumer app

The arena’s verdict

Pick Grok 4.3 if you run agentic or high-volume pipelines where cost per call dominates: it delivers near-frontier reasoning and a big tool-calling jump over Grok 4.20 at a fraction of GPT-5.5 or Gemini 3.1 Pro pricing. Skip it if coding precision is your priority, as developers still rank Claude and the big April releases ahead. Also stay on Grok 4.20 if you need its 2M context or its better non-hallucination score for legal, medical, or compliance work. Latency-sensitive apps should test the ~13s time to first token before committing.

Thumbs up or thumbs down

Cast your verdict

Would you recommend Grok 4.3, or warn the crowd away?

50%crowd score · 0

Top Grok 4.3 alternatives

All alternatives
1
Claude Haiku 4.5

Anthropic's fastest model: about 90% of Sonnet 4.5's coding skill at $1/$5 per 1M tokens, 200K context.

$5/1M out67%(2)
Visit
2
Claude Opus 4.7

Anthropic's April 2026 Opus: 87.6% SWE-bench Verified, 1M context, high-res vision, now behind Opus 4.8

$25/1M out57%(3)
Visit
3
Claude Opus 4.8

Anthropic's flagship Opus-tier model for long-horizon agentic coding; 1M context at $5/$25 per 1M tokens.

$25/1M out57%(3)
Visit

Compare Grok 4.3 head-to-head

Grok 4.3: frequently asked questions

How much does Grok 4.3 cost per 1M tokens?

Grok 4.3 costs $1.25/1M in per 1M input tokens and $2.50/1M out per 1M output tokens. Flat $1.25 in / $2.50 out across all reasoning-effort tiers; cached input $0.20/1M. xAI's own pricing page shows no long-context surcharge, but OpenRouter lists tiered higher rates above 200K total tokens.