Grok 4.6

xAI's news index confirms the August 12, 2026 launch of Grok 4.6 and enumerates a sequence of platform distribution milestones in the weeks that followed. Grok 4.6 was added to GitHub Copilot as xAI's "latest coding model" on August 14, 2026, and subsequently became available via Amazon Bedrock on August 19, 2026, givi Further distribution entries show Grok 4.6 arriving on the Gemini Enterprise Agent Platform (entry dated August 21, 2026) and on Microsoft Foundry (entry dated August 26, 2026), with each item explicitly naming Grok 4.6 as the model being made available. The same index also lists product-level news such as Grok Build r

On August 12, 2026, xAI released Grok 4.6, positioning it as an evolution of Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work. According to the official announcement page, the model is designed to stay with complex tasks across many steps, whether researching a topi The announcement documents a training approach that builds on Grok 4.5 with a longer supplemental run using curated model-generated reasoning data, high-quality engineering data, and an improved optimizer; SFT trajectories were regenerated with Grok 4.5 across reasoning efforts and agent harnesses, then refined through

IntuitionLabs's September 5, 2026 analysis characterizes Grok 4.6 as a coding- and agent-focused frontier model with a 500,000-token context window, configurable reasoning effort, native function calling, and API pricing of $2 per million input tokens and $6 per million output tokens. Independent measurement from Artif The analysis highlights that no public benchmark yet evaluates Grok 4.6 specifically against research-software repositories, and recommends that research software teams weigh its price and general agentic strength against documented terminal-tool-use weak points. Pricing tiers cited from xAI documentation show $2/$0.50

Kie.ai's August 13, 2026 same-day analysis confirms Grok 4.6 scored 61 on the Artificial Analysis Intelligence Index, up 5 points from Grok 4.5 and +23 from Grok 4.3, matching GPT-5.6 Sol and sitting two points behind Claude Opus 5 while holding pricing flat at $2/$6 per million tokens. The piece notes the headline "ma The analysis observes that no independent third-party replication of the numbers existed as of August 13, 2026, and frames Grok 4.6's main selling point as price-to-intelligence rather than raw leadership. SpaceXAI's official announcement described the model as building on Grok 4.5 with a focus on long-running agents a

The AIToolsReview piece reports independent Artificial Analysis evaluation placing Grok 4.6 at 61 on the Intelligence Index, tying GPT-5.6 Sol and trailing Claude Fable 5 and Claude Opus 5 by one and two points, a five-point gain over Grok 4.5. It highlights efficiency: Grok 4.6 resolves tasks in roughly half the turns It also flags practical caveats: a 200,000-token pricing cliff that reprices an entire long request once crossed, and an independently-measured confident-fabrication rate that the authors say is meaningfully higher than its benchmark parity with GPT-5.6 Sol might suggest. The review treats Grok 4.6 as SpaceXAI's fourth

Artificial Analysis's independent evaluation places Grok 4.6 at 61 on the Intelligence Index, a 5-point gain over Grok 4.5 and +23 over Grok 4.3, joining the frontier in line with GPT-5.6 Sol, behind Claude Opus 5 (63) and Claude Fable 5 (62), and just ahead of Kimi K3. The model scored a GDPval-AA v2 Elo of 1753, behi Context window remains 500k tokens unchanged from Grok 4.5, and headline pricing holds at $2 per million input tokens and $6 per million output tokens, which Artificial Analysis describes as 60%+ below Claude Opus 5 ($5/$25) and GPT-5.6 Sol ($5/$30). Cost per task was measured at $0.84, the same as Kimi K3 but with sli

This DEV Community write-up documents Grok 4.6's training-recipe changes: a longer supplemental training run using curated model-generated reasoning and technical data, a supervised fine-tuning stage where Grok 4.5 itself regenerated SFT trajectories across reasoning, agent harnesses, STEM, software engineering, and kn Behaviorally, the post says Grok 4.6 shows more self-testing and verification on longer trajectories and produces stronger first passes on visual and interactive projects than Grok 4.5. It restates the official headline that Grok 4.6 matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index composite, with pric

Snorkel AI's independent Senior SWE-Bench evaluation (published September 3, 2026) places Grok 4.6 second on basic solve rate at 65.3% pass@3, behind Claude Opus 5 at 70.5%. On stricter tasteful solve rate it ranks fifth at 38.9%, and on pass^3 (all three attempts succeeding) it ranks fourth at 16.8%. The benchmark was developed with Princeton and UW-Madison researchers. The standout result is cost-performance: Grok 4.6 averaged 23.9K output tokens per task, roughly $0.14 in output cost, versus $1.77 for Opus 5 and $2.92 for Claude Fable 5 on the same measure. All models ran on the Mini-SWE-Agent harness via Harbor with three attempts per task, scored over 95 tasks in the current 100-task set. The leaderboard snapshot is dated August 20, 2026.

Better Stack's August 31, 2026 community guide reports that Grok 4.6 was released on August 12, 2026, 35 days after Grok 4.5, with pricing unchanged at $2 per million input tokens and $6 per million output tokens, described as the cheapest model currently competing at the intelligence frontier. The model scores 61 on t The guide notes that xAI has no published architecture details for Grok 4.6: no parameter count, no MoE disclosure, and no system card, with reports of a 1.5-trillion-parameter base coming from third-party analysis rather than vendor statements. Confirmed improvements are in post-training, including upgraded supervised