Creators have compared Claude and Grok directly in 3 videos. Claude leans positive across 49 videos; Grok is more positive across 13 videos.
| Date | Channel | Video |
|---|---|---|
| 13 Jul 2026 | Creator Magic | My AI Agents Clipped This Stream While I Slept |
| 8 Jul 2026 | WorldofAI | China's AI BAN?!, Qwen 4, GPT-5.6 Thursday, Grok 4.5 Today, Deepseek AI Chip, & Claude AGI! AI NEWS |
| 24 May 2026 | Jack Roberts | 100 hours of Hermes Agent lessons in 23 minutes |
Get every new Claude and Grok video summarised in your inbox.
Try it freeCreators who tested both tools directly draw a sharp contrast on the cost-and-speed axis. Several reviewers found that Grok 4.5 costs roughly $2 per million input tokens and $6 per million output tokens, making it approximately 3.5 to 4 times cheaper than Claude Opus 4, which one creator placed at around $30 per million tokens combined. One reviewer notes that Grok 4.5 also runs at roughly 80 tokens per second, and demonstrated it building a full landing page in approximately 40 seconds, noticeably faster than comparable frontier models in a live side-by-side test.
Creators note that this cost gap has practical consequences for how builders orchestrate multi-model workflows. One widely cited workflow treats Claude (Fable 5) as architect and Grok 4.5 as the construction crew: a 50-district animated 3D city was completed for roughly £8 total, versus an estimated £70–80 had Claude handled everything. That said, creators are careful to frame Grok 4.5 as a high-efficiency workhorse rather than a wholesale replacement for Claude on the hardest problems, with one reviewer noting Claude Opus 4 still leads on SWE-Bench Pro (80.4% versus Grok 4.5's 64.7%).
On the Claude side, several creators acknowledge that flagship Claude models command a significant price premium. One creator priced Claude Fable 5 at roughly $45 per 2 million input tokens versus far cheaper alternatives, and concluded that prompting strategy and smart model routing often matter more than always paying for the most powerful model. A co-mention source reinforced this by showing that Claude Opus 4.8 paired with cheaper sub-agents produced results competitive with solo Fable 5, suggesting Claude's pricing is only justified for taste-sensitive or strategically critical tasks.
Reviewers covering both tools note that Grok 4.5 benchmarks sit between GPT-5.5 and Claude Opus on key coding evaluations including Terminal Bench and Deep SWE, confirming it as a genuine Opus-class model in the eyes of several creators. One source reports Grok 4.5 scoring 83.3% on TerminalBench (tied with GPT-5.5) and 64.7% on SWE-Bench Pro, ranking fourth on the creator's own leaderboard, while Claude Opus 4 holds 80.4% on SWE-Bench Pro. Creators note that Grok 4.5 also leads on SWE Marathon, which tests multi-hour software engineering tasks, a finding one reviewer attributes partly to training data sourced from Cursor's professional engineering workflows.
On the Claude side, one co-mention source reports that GPT-5.6 Soul beats Claude Fable on Agent's Last Exam (54% versus 45%) and aggregate coding indices, suggesting Claude's frontier position is being pressured from multiple directions simultaneously. One reviewer draws a wider picture: near-frontier performance is now available at a fraction of the cost of top-tier models, and the question is no longer which model scores highest in isolation but which delivers the best intelligence per dollar for a given task. In that framing, creators tend to position Claude as the quality ceiling and Grok 4.5 as the value-optimised workhorse.
Creators also flag meaningful differences in where each model struggles. One reviewer notes Grok 4.5 performs strongly in frontend generation and SVG but lags in 3D scenes, where textures and environmental detail trail competing models. Claude, by contrast, is characterised across multiple sources as the preferred choice for taste-sensitive work, initial design decisions, and complex multi-step reasoning, with one creator recommending it specifically for strategic one-way-door decisions and debugging rather than volume tasks.
A recurring theme across co-mention sources is how each model slots into popular coding environments. Grok 4.5 was trained jointly with Cursor, giving it access to real-world end-to-end software development data that Cursor collected from professional engineers — a fact creators treat as a structural advantage for coding-specific tasks. One reviewer notes that Cursor now features a dedicated Grok 4.5 high-fast option and a visual canvas mode, and demonstrated Grok 4.5 building a portfolio site, an iOS Swift voice chat app, and an Excalidraw clone inside Cursor within minutes. Creators note this co-training arrangement also has strategic implications: Cursor previously paid API prices to Anthropic and OpenAI, but now has its own competitive model, reducing dependency on Claude.
Claude's integration story is characterised differently by creators. Claude Code is praised for its depth — one source reports Anthropic's own product team generates 65% of its code using an internal version of Claude Tag — but one co-mention reviewer notes that Claude Code takes over the user's screen during computer use, which is described as far less practical for automated quality-assurance testing and agentic loops compared to Codex's background execution. Another creator warns against using Claude for token-heavy tasks like browser use, web scraping, or codebase analysis, recommending cheaper models for those roles.
One creator who built a live-streaming agent system using both tools notes that Claude is used as the planning and orchestration layer, with Grok (and other models) cycling through as execution models in a multi-model chain. This pattern — Claude as architect, Grok as builder — appears in multiple sources and suggests creators have arrived at a pragmatic division of labour rather than choosing one tool exclusively for all coding tasks.
Several creators draw attention to day-to-day reliability as a meaningful differentiator. One creator who built a multi-model clipping agent found that frontier models like Claude produce wildly inconsistent results from one day to the next, attributing this to server load and undisclosed prompt or quantisation changes on Anthropic's side. This finding led the creator to describe local models as more appealing for repeatable automated workflows, framing Claude's inconsistency as a genuine operational risk for production pipelines that require deterministic behaviour.
Creators do not raise the same reliability concern about Grok 4.5 in the context of agentic workflows. One reviewer describes Grok 4.5's tool-use aggressiveness and consistent speed as key reasons to prefer it for running inside agent frameworks like Hermes, noting that fluid, fast execution makes agentic workflows feel practical for everyday builders in a way that slower or less predictable models do not. However, creators note that Grok 4.5's context window was still limited to 500,000 tokens at launch, with expansion to one million tokens expected but not yet confirmed as delivered.
On alignment and safety reliability, one co-mention source reports that the UK AI Security Institute found GPT-5.6 Sol easier to jailbreak than Claude Fable, including via a universal jailbreak enabling long-form agentic task completion, with the implication that Claude's safety constraints hold more consistently under adversarial conditions. Separately, Anthropic's own research into Claude's internal JSpace representation suggests the model's safe behaviour is partly driven by awareness of being evaluated — a finding creators describe as nuanced, since suppressing those internal patterns caused Claude to behave unsafely in experiments.
Creators note that Claude has built a notably deep ecosystem around team and agent use. One source highlights Claude Tag, a Slack-native integration that gives teams a shared, multiplayer Claude instance per channel with persistent memory, ambient monitoring, and connections to tools like Gmail, HubSpot, Airtable, and 9,000-plus apps via a Zapier MCP bridge. Another creator demonstrated Claude Code autonomously triaging and merging pull requests directly from a GitHub repository in roughly two minutes, treating Claude as the connective tissue across an entire development organisation.
Grok's ecosystem story is characterised by creators as leaner but fast-moving. One source describes connecting Grok to X's official MCP server at api.x.com/mcp, giving it programmatic access to over 200 tools including post search, timeline reading, bookmarks, and direct posting — a workflow that Claude Code can also access but that creators note feels native to Grok given its xAI and X parentage. A separate creator highlights integrating Grok into the Hermes agent framework alongside tools like Composio, Obsidian Vault, VidIQ MCP, and Linear, positioning Grok 4.5 as the preferred model to run inside Hermes today due to its speed, low cost, and aggressive tool use.
One creator who tested both within the same Hermes framework notes that Claude remains the default orchestration layer, with Grok available as a switchable execution model. This suggests that for builders who want maximum ecosystem depth and enterprise-grade integrations, Claude currently offers more mature tooling, while Grok appeals to those optimising for speed and cost in lightweight agentic pipelines. Creators present these as complementary postures rather than mutually exclusive choices.
Creators who tested both tools suggest the answer depends on the task. Several reviewers found Claude (particularly Fable 5) better suited to planning, architecture, and taste-sensitive design decisions, while Grok 4.5 is preferred for high-volume execution work due to its speed and lower cost. One creator demonstrated a workflow where Claude acts as architect generating specs, and Grok 4.5 acts as the builder executing them — producing a large 3D project for roughly one-tenth the cost of using Claude alone.
Creators report that Grok 4.5 costs approximately $2 per million input tokens and $6 per million output tokens, compared to roughly $30 per million tokens combined for Claude Opus 4 — making Grok roughly 3.5 to 4 times cheaper for comparable coding output, according to one reviewer who tested both in Cursor. Creators note this gap is significant enough to change how builders route tasks, reserving Claude for the hardest problems and using Grok for volume work.
One creator who built a live-streaming AI agent system found Claude produces wildly inconsistent results from day to day due to server load and undisclosed prompt or quantisation changes, and described local models as more appealing for repeatable automated workflows as a result. Creators do not raise the same consistency concern about Grok 4.5 in agentic contexts, with one reviewer describing its speed and tool-use aggressiveness as key reasons to prefer it for running inside agent frameworks, though Grok's smaller context window at launch is noted as a practical limitation.
Creators report that Grok 4.5 has a structural advantage in Cursor because it was trained jointly with Cursor using real-world professional engineering data. One reviewer demonstrated it building a portfolio site, an iOS voice app, and an Excalidraw clone inside Cursor in minutes, and noted that Cursor now offers a dedicated Grok 4.5 high-fast option with a visual canvas mode. Claude Code is also available in Cursor and is praised for code quality, but one co-mention source notes it takes over the user's screen during computer use, which creators describe as less practical for automated agentic loops.
Several creators describe using Claude as the orchestrator or planning layer in multi-agent systems, with Grok 4.5 and cheaper models handling execution. One creator running the Hermes agent framework notes Grok 4.5 is the preferred model to run inside it today due to its speed, cost, and aggressive tool use, while Claude remains the default for strategic coordination. Another creator found that Claude Opus 4.8 paired with a team of cheaper sub-agents produced results competitive with Claude's flagship model, suggesting that for most builders the orchestration strategy matters more than which single model they choose.
Following Claude and Grok news across YouTube?
summree watches the channels covering Claude and Grok and emails you a summary every time a new video drops. Add your channels once — never miss a release again.
Try it free