Grok 4.6: SpaceXAI Matches GPT-5.6 Sol at a Fraction of the Price

SpaceXAI (formerly xAI) released Grok 4.6 on August 12, 2026, and the numbers immediately shook the AI landscape. The model scores 61 on the Artificial Analysis Intelligence Index — exactly tying OpenAI's flagship GPT-5.6 Sol Max while costing roughly 60% less on input and 80% less on output . For developers and enterprises watching their inference bills, this is the biggest pricing shock in a year that has already seen OpenAI cut Luna by 80% and DeepSeek launch a 99%-cheaper alternative . Grok 4.6 is not a new foundation model — it is a post-training upgrade on the same 1.5-trillion-parameter V9 base as Grok 4.5 — but the improvements in agentic coding, knowledge work, and multi-step reasoning are enough to put it back on the frontier .

The Announcement and Key Specs

SpaceXAI positioned Grok 4.6 as its frontier model for coding, long-running agents, and knowledge work, with specific tuning for multi-step tasks and interactive/visual projects . It shipped on August 12, 2026 across the SpaceXAI API, Cursor, Grok Build, Grok Bot, OpenRouter, Vercel, and Cloudflare on day one, with 2x included usage in Cursor and Grok Build for the first week .

The model ID is grok-4.6, and the key documented specs are :

• Context window: 500,000 tokens
• Knowledge cutoff: February 1, 2026
• Modalities: Text and image input; text output
• Output limit: No stated limit
• Reasoning levels: Low, medium, high (default), xhigh
• API price (input): $2.00 per 1M tokens (below 200K prompt tokens)
• API price (cached input): $0.50 per 1M tokens
• API price (output): $6.00 per 1M tokens
• Long-prompt pricing (≥200K tokens): $4.00 / $1.00 / $12.00 per 1M tokens

How Grok 4.6 Compares to Competitors

Model	Company	Release	Input $/1M	Output $/1M	Context	AA Intelligence Index
Grok 4.6	xAI/SpaceXAI	Aug 12	$2.00	$6.00	500K	61
GPT-5.6 Sol	OpenAI	Jul 9	$5.00	$30.00	~1.05M	61
Claude Fable 5	Anthropic	~Aug 2026	$10.00	$50.00	1.0M	62
Claude Opus 5	Anthropic	Jul 24	$5.00	$25.00	1.0M	63

Sources:

The headline numbers:

• Grok 4.6 ties GPT-5.6 Sol on the overall Intelligence Index (61) while costing 60%+ less on input and 80% less on output .
• Sits 1 point behind Claude Fable 5 (62) and 2 points behind Claude Opus 5 (63), at a fraction of the cost .
• Cache-hit pricing of $0.50 per million input tokens is industry-leading for a frontier model .

Where Grok 4.6 wins:

• Agent efficiency: It resolves tasks in roughly 53 turns vs. ~103 turns for Claude Opus 5, using a quarter of the input tokens .
• Cost per task: Approximately $0.84 per completed task on Artificial Analysis's index, the lowest among frontier models .
• Specific benchmarks: On GDPVal-AA v2 (knowledge work), it scores 1,753 — ahead of GPT-5.6 Sol Max (1,728) and Claude Fable 5 (1,741) . On OfficeQA Pro, it scores 63.2%, matching or exceeding Sol on professional-work evals .

Where it trails:

• Terminal-Bench v3.0: 26% vs. 34.6% for GPT-5.6 Sol Max and 34.1% for Fable 5 Max .
• Context window: 500K tokens vs. ~1.05M for GPT-5.6 Sol and 1M for Claude Opus 5 .
• Multimodal: Text output only; no multimodal generation .

Strengths and Limitations at a Glance

Strengths

• Frontier-level intelligence (61 AA Index) at 2–10x lower API cost than rivals
• Leading agent efficiency: half the reasoning turns of Claude Opus 5
• Lowest cost per task among frontier models (~$0.84)
• Strong on agentic coding and knowledge-work benchmarks

Limitations / Unknowns

• Not a new foundation model — gains come entirely from better SFT and RL on the same V9 base as 4.5
• Smaller context window (500K) vs. GPT-5.6 Sol (~1.05M) and Claude Opus 5 (1M)
• Text-only output; no image, audio, or video generation
• Independent third-party replication of benchmark numbers was not yet available as of August 13, 2026

Google's Response on August 13

The available evidence does not surface a specific Google or Gemini announcement from August 13 in direct response to Grok 4.6. No significant counter-move from Google DeepMind was captured in the search results as of this writing.

Broader Context: The AI Price War Timeline

Grok 4.6 enters a market that has seen relentless price compression:

• July 8, 2026 — SpaceXAI launched Grok 4.5 at $2/$6 per 1M tokens .
• July 9, 2026 — OpenAI shipped the GPT-5.6 family (Sol, Luna, Terra), with Sol at $5/$30 .
• July 17, 2026 — Meta entered the price war with a model priced at ~25% of OpenAI/Anthropic rates .
• July 24, 2026 — Anthropic released Claude Opus 5 ($5/$25) and Claude Fable 5 ($10/$50) .
• July 30, 2026 — OpenAI slashed GPT-5.6 Luna by 80% (to $0.20/$1.20) and Terra by 20%, citing token-generation efficiency gains rather than a price war .
• August 1, 2026 — DeepSeek launched a model charging ~$0.28 for output that costs $25 on Opus 4.8 — a 99% discount .
• August 12, 2026 — Grok 4.6 holds price flat from 4.5 while jumping 5 points on the Intelligence Index .

The competitive dynamic has shifted: frontier labs are no longer competing only on benchmark scores (which have largely converged), but on cost per task and agent efficiency. Grok 4.6's core play is offering frontier-level capability at 2–10x lower cost than its closest rivals, putting pressure on Anthropic and OpenAI's inference margins ahead of their planned IPOs .