Grok 4.6 vs Grok Build 0.1: General AI, Coding Agents, Context, and Speed
Grok 4.6 and Grok Build 0.1 are xAI's two current flagship-adjacent models, and they are not competing options — they were built for different jobs and priced accordingly.
........
Grok 4.6, released August 12, 2026, is xAI's general frontier model: 500K context, $2 input and $6 output per million tokens.
Grok Build 0.1 is the coding specialist: 256K context, $1 input and $2 output per million tokens, the cheapest model xAI publishes.
Grok 4.6 has configurable reasoning effort including a true xhigh level; Grok Build 0.1's reasoning is always on and cannot be tuned.
Grok 4.6 crosses a long-context pricing cliff at 200K input tokens, doubling rates for the entire request.
On the Artificial Analysis Intelligence Index, Grok 4.6 scores 61, roughly level with GPT-5.6 Sol.
··········
TWO MODELS, TWO JOBS.
The clearest way to think about this pair is that only one of them is a general-purpose model.
Grok 4.6 is xAI's current frontier model, designed for long-running agents, coding, research, and interactive or visual work. It accepts text and image input and returns text.
Grok Build 0.1 is a coding model trained specifically for agentic software engineering — refactoring, tool invocation, structured outputs, and multi-step development loops. It also accepts text and images, which lets source code, diagrams, UI mockups, and error screenshots go into the same request.
Grok Build 0.1 additionally underpins the Grok Build CLI, xAI's terminal-native coding agent, though the model is available through the API independently of that product.
So the question is rarely which is better. It is whether the work is software engineering, in which case Build 0.1 is the specialist, or something broader, in which case 4.6 is the general model.
··········
PRICING.
Grok Build 0.1 is the cheapest model xAI publishes, and the gap against 4.6 is wide on output in particular.
Grok 4.6 lists at $2.00 per million input tokens, $0.50 per million cached input tokens, and $6.00 per million output tokens, for prompts below 200,000 tokens.
Grok Build 0.1 lists at $1.00 per million input tokens, $0.20 cached, and $2.00 per million output tokens.
On output — the token class that matters most for agentic coding, which generates far more than it consumes — Build 0.1 costs a third of 4.6.
For context on where Grok 4.6 sits in the wider market, $6.00 output puts it below the frontier-tier alternatives it competes with, several of which sit at $10, $15, $25, and $30. That output pricing is a significant part of why the model appeared in coding agents early: those workloads are output-heavy.
··········
Head-to-head specifications
Grok 4.6 | Grok Build 0.1 | |
|---|---|---|
Released | August 12, 2026 | May 2026 |
Context window | 500K tokens | 256K tokens |
Input per 1M | $2.00 | $1.00 |
Cached input per 1M | $0.50 | $0.20 |
Output per 1M | $6.00 | $2.00 |
Reasoning control | Configurable, incl. xhigh | Always on, not configurable |
Input types | Text and image | Text and image |
Long-context tier | Doubles at 200K | Not published |
··········
THE 200K CLIFF ON GROK 4.6.
Grok 4.6's headline rate applies only below a threshold, and crossing it is expensive in a way that catches people out.
Once a prompt reaches 200,000 input tokens, xAI bills the entire request at $4.00 per million input, $1.00 cached input, and $12.00 output — double the standard rates, applied to the whole request rather than just the portion above the line.
With a 500,000 token context window, this is not a theoretical edge case. Any workload that routinely loads large repositories or long document sets will spend much of its time above the threshold.
The practical implication: the framing of Grok 4.6 as roughly half the price of frontier competitors does not hold for long-context workloads. Anyone whose prompts run large should model the long-context tier before assuming the cheaper positioning applies to them.
··········
A CACHING CHANGE WORTH NOTICING.
Grok 4.6 kept the same headline rates as Grok 4.5 but raised the price of exactly the token class agent loops consume most.
Grok 4.5, released July 8, 2026, charges $0.30 per million cached input tokens. Grok 4.6 charges $0.50 — a 67% increase on cached input, while standard input and output rates stayed at $2 and $6.
Reports also indicate Grok 4.6 is measurably more verbose than 4.5, which means the same task generates more billable output tokens even though the output rate did not move.
Both effects push in the same direction: for a stable-prefix agent workload, migrating from 4.5 to 4.6 can raise the bill despite an unchanged headline price.
Grok 4.5 remains available with the same 500K context window, so existing applications are not forced to move. The reasonable approach for new deployments is a controlled comparison on representative tasks, since a newer model does not automatically produce a lower total cost.
··········
REASONING CONTROL.
This is a genuine architectural difference rather than a pricing one.
Grok 4.6 supports configurable reasoning effort, including a true xhigh level. xAI notes that an xhigh request sent to Grok 4.5 is treated as high, so the top setting is specific to 4.6.
Grok Build 0.1 has reasoning always active with no configurable effort and no non-reasoning mode. Every response includes structured analysis before the final output.
For a coding agent that is arguably the right default — the loop depends on the model reasoning through errors rather than pattern-matching a fix. But it removes a cost lever: there is no way to dial reasoning down for simple edits.
On Grok 4.6, effort configurability is a real budgeting tool, since reasoning tokens are billed as output at $6 per million below the long-context threshold.
··········
BENCHMARK POSITION.
Grok 4.6 lands roughly level with the top of the current field on independent aggregate scoring.
On the Artificial Analysis Intelligence Index, Grok 4.6 scores 61, putting it approximately level with GPT-5.6 Sol and around five points above Grok 4.5.
In the same analysis, Claude Opus 5 scored 63 and Claude Fable 5 scored 62 — placing Grok 4.6 just below both, while undercutting them substantially on price.
Grok Build 0.1 is harder to place on the same scale because it is a specialist model measured against different criteria. Independent cost-per-task benchmarking for Grok 4.6 was still pending at the time of writing at at least one benchmark operation, which is a reason to treat any single ranking as provisional.
··········
WHERE YOU CAN ACTUALLY GET THEM.
Availability differs by route, and one gap is significant for enterprise buyers.
Grok 4.6 is live through the xAI API, Grok Build, Cursor, Grok Bot, OpenRouter, Vercel, and Cloudflare. New API requests use the model ID grok-4.6.
Azure AI Foundry and AWS Bedrock are still a generation behind — Azure's Grok pricing tops out at Grok 4.3 at $1.25 input and $2.50 output. That is cheaper per token and two releases old. Any organisation whose procurement mandates inference through one of those clouds cannot buy Grok 4.6 at any price yet.
On OpenRouter, the listed rates match xAI's exactly, with OpenRouter stating it passes through provider pricing without markup. The cost sits elsewhere: a 5.5% fee on credit purchases with an $0.80 minimum, 5% on crypto, unused credits expiring after a year, and a 5% fee on usage above plan allowance when bringing your own xAI key.
Grok Build 0.1 is available through the xAI API in public beta with no subscription required, and the Grok Build agent product is bundled into SuperGrok and X Premium Plus subscriptions.
··········
CONSUMER PLANS ARE A SEPARATE BILL.
Two very different questions hide behind the phrase Grok pricing, and conflating them is the most common mistake.
The API is metered per token. The consumer subscription is a monthly fee. They are separate products with separate bills, and a SuperGrok subscription does not include API credits.
As of mid-August 2026, xAI listed Grok 4.6 with the $30 per month SuperGrok plan, and SuperGrok Plus at $100 per month, which adds significantly higher usage across Chat, Imagine, Voice, and Build, 1080p video creation, faster replies, peak-time priority, and early access. SuperGrok Heavy sits around $300 per month.
Paid Grok plans now run on a single shared weekly usage pool spendable across Chat, Imagine, Voice, and Build — the per-two-hour prompt caps that older guides still quote were retired in June 2026.
xAI's usage view can attribute a plan's weekly allowance across API, Build, Chat, Imagine, and Voice, but xAI does not describe that as a separately issued developer credit balance; developer requests remain metered at API rates.
··········
CHOOSING BETWEEN THEM.
The decision comes down to workload shape rather than to which model is stronger.
Use Grok Build 0.1 when the work is software engineering and token cost matters: agentic coding loops, refactoring, debugging, web development, and MCP-based tool workflows. It is a third the output price and purpose-trained for the task.
Use Grok 4.6 when the workload benefits from the newest general model and sustained multi-step execution: demanding coding agents that also need broader reasoning, multi-step research workflows, or interactive and visual applications.
Use Grok 4.3 instead of either when token cost dominates and having the latest frontier model does not matter — it sits at $1.25 input and $2.50 output with a 1M context window.
Whichever direction the analysis points, the safest migration is a controlled comparison on representative tasks rather than a global switch, particularly given how much the long-context tier and the cached-input rate can move a real bill.
··········
·····
FOLLOW US FOR MORE.
·····
·····
DATA STUDIOS
·····
