AI Coding Agent Pricing in 2026: Billing Models, Cost Variables, and How to Budget

AI coding agent pricing is not the per-seat number on the pricing page. That is the central claim of a detailed guide TrueFoundry published on August 6, 2026 — and it matches what agencies see when their clients' first coding-agent invoices land. Flat, credit, and pay-per-token plans respond differently to the same usage, and model choice can move the bill more than seat count does.

This page distills the guide into the four billing models — including Grok Bot's new subscription-bundled agent access — the seven cost variables that predict spend, what pricing pages omit, and a six-step budgeting checklist you can hand to finance. It also covers Slack Code, the agentic-coding channel Salesforce launched August 20, 2026 that makes Claude, GitHub Copilot, Devin, and Vercel agents accessible from collaborative Slack channels. If you need to price agency delivery of AI-assisted work rather than internal seats, we cover that too — including what the August 2026 DHH–Lex Fridman debate on vibe coding vs agentic engineering means for projects vs retainers. And if you're choosing between Claude Code and Codex CLI — or deciding which harness your agency should standardize on — the August 2026 agent harness rankings are below.

The four AI coding agent billing models

Every major tool — Cursor, GitHub Copilot, Windsurf, Claude, OpenAI Codex, and now Grok Bot — maps to one of four billing structures, per TrueFoundry's August 6, 2026 guide:

Grok Bot also breaks the tool-interop assumption. SpaceXAI says its bots can sign into and work across apps, tools, and websites — including platforms with no clean API or MCP — while Claude Code, OpenAI Codex, and Muse Code rely on MCP (and documented APIs) for tool interop. That matters for the budget conversation: an agent that works where the work lives can replace more manual steps, but with no standalone price and undisclosed authentication mechanics, agencies should treat it as a beta evaluation, not a budget line item.

"Pricing pages show the entry point. Invoices show consumption. Budget for the second, not the first." — TrueFoundry, AI Coding Agent Pricing (Aug 6, 2026)

Slack Code: AI coding agents, now inside collaborative Slack channels

On August 20, 2026, Salesforce launched Slack Code — dedicated "code channels" inside Slack where teams and AI coding agents build software together. Slack frames it as "multiplayer" agentic coding: mention an agent like Claude or Devin in a conversation, and it spins up a project-specific code channel, pulling in relevant teammates and context. The channel shows code diffs, live HTML previews, and planning docs; anyone in the channel can review, give feedback, pause, redirect, or stop the agent, and approve work before it ships. When the work is done, the channel archives itself while the history stays searchable, like an audit log.

Availability: Slack Code is available on any Slack plan from day one, with four agent integrations live at launch: Claude (Anthropic), Devin (Cognition), GitHub Copilot, and Vercel. OpenAI's ChatGPT is a founding partner but was "available soon" at launch — the only named agent not yet usable in code channels. Slack says more partners are on the way, and it is pitching code channels beyond engineering: marketing campaigns, legal document review, IT onboarding.

Pricing impact: Slack Code itself has no separate price — it ships inside Slack plans. But per Salesforce, "access to each partner agent is required": agencies still pay for Claude, Copilot, Devin, or Vercel subscriptions on top of Slack seats. It is a distribution layer, not a new billing model — the four billing structures above stay valid; what changes is where agent work happens and who sees it.

Why it matters for agency decision-making:

• Distribution shifts from IDE/terminal to the collaboration platform. Coding agents that used to live in a developer's local session are now accessible through collaborative Slack channels. Slack is competing to be where agentic coding happens, alongside the established IDE and CLI channels.
• Slack becomes a client-facing delivery surface. Code channels expose the whole build — diffs, previews, decisions — to everyone in the channel. Clients can watch a fix happen and approve it in-channel instead of waiting on status meetings. Slack's pitch: "No ticket, no meeting, no waiting — just a fix, shipped."
• Human sign-off is the compliance story. High-stakes moves, like pushing code to production, are packaged for an expert to sign off on right in the channel, governed by Slack's existing enterprise security model. Agencies can position this as a sellable guardrail: the agent proposes, your engineer approves, the client sees it happen.
• Budget for both lines. Slack Code does not bundle agent subscriptions. When scoping client work, model Slack plan seats and separate agent subscriptions as distinct cost lines.

Analysis: the implications above are editorial inference from the sourced launch facts, not claims made by the sources. Slack Code does not change what coding agents cost; it changes where agent work happens — and who sees it. Teams already on Slack gain a zero-marginal-cost surface for agent delivery; the marginal cost remains the per-agent subscription, which the pricing math above still governs.

See our Slack Code launch brief for full coverage of what agencies should do next.

Claude Code weekly limits change September 14: +25% permanent baseline, net −17% vs today

On August 29, 2026, Anthropic announced that starting September 14, 2026 it is permanently raising Claude Code's standard weekly usage limits by 25% over the pre-promotion baseline for Pro, Max, Team, and seat-based Enterprise plans — and the temporary 50% weekly boost (active since May 13, 2026, extended through the summer) expires the same day (@ClaudeDevs, Aug 29, 2026; BleepingComputer). Anthropic confirmed the net effect itself: "Compared to today, this works out to a 17% reduction in weekly limits on Claude Code."

Going from 150 to 125 is a 16.7% cut (25 ÷ 150), which Anthropic and every major outlet round to about 17%. A workload consuming 100% of today's boosted cap will consume 120% of the new cap (150 ÷ 125 = 1.2) starting September 14.

Plan scope: the change applies to Pro, Max (5x and 20x), Team (standard and premium seats), and seat-based Enterprise plans. Free and consumption-based (pay-as-you-go) Enterprise seats are excluded. It is a weekly-capacity change only — five-hour session limits, plan pricing, Claude chat, Claude Cowork, and Console API pay-per-token pricing are unchanged. For Pro and Max, Claude apps and Claude Code draw from the same weekly bucket; Team plans give individual member limits, not an org-wide pool.

One documentation note: as of August 30, 2026, Anthropic's Help Center promo article still lists the 50% boost as running through August 31 at 11:59 PM PT; the September 14 date comes from Anthropic's announcement on X. Plan on the September 14 schedule, since it is the company's stated date.

For agencies running Claude agents at scale, three implications:

• Re-baseline capacity at 125, not 150. If a client delivery workflow only fits at 150% of the standard weekly quota, it hits a hard wall when the permanent 125% level takes over on September 14. Model headcount against the post-change cap and treat anything above it as headroom.
• Quota economics are the cost story, not the sticker price. The weekly pool is shared across Claude Code and the Claude apps; Opus-class models burn several times more per turn than Sonnet; extended thinking and subagent fan-out accelerate burn; and there is no published numeric cap for any plan — /usage in the CLI is the only ground truth.
• September 13 is the last day of the boost. If you lean on Claude agents for client delivery, front-load spike work (release crunches, migrations) before the ceiling drops, and quote capacity at the new 125% baseline with the +50% treated as time-boxed runway.

See Claude Code limits drop 17% on Sept 14 — what agencies should do for the full breakdown and the five pre-change actions.

The seven cost variables that predict spend

TrueFoundry's guide argues six factors predict spend better than the headline number — and OpenAI's August 11, 2026 agent-import wave added a seventh: migration and import effort. Together:

What pricing pages tend to omit

• Token quotas on free and entry tiers, often too low for daily professional use.
• Overage behavior on paid tiers, sometimes buried or unpublished until you hit a limit.
• Model-specific credit consumption rates — the model picker is a cost lever; pricing pages rarely say so.
• Enterprise and custom pricing that makes comparison impossible without a sales conversation.

The working example in the source is stark: a task that costs one credit unit on a standard model might cost eight on a frontier model. Budget against that spread, not against the cheapest option on the pricing page.

The coding-agent price war is starting to bend both spread variables: Meta launched Muse Code (beta, no GA date confirmed) on Aug 5, 2026 at $1.25/M input and $4.25/M output — undercutting Claude Code API and OpenAI Codex at volume — so the 20–50× usage spread and the 8× credit-burn rate become bigger levers in the budget conversation when teams switch models to chase cheaper tokens.

Budgeting for AI coding agents: a six-step checklist

1. Map your team's usage profile. Headcount first, then intensity: light (autocomplete, occasional prompts), moderate (daily agent-assisted tasks), or heavy (continuous agent workflows across large codebases). Model a range if uncertain — the gap between low and high estimates is your budget risk exposure.
2. Match the billing model to your priorities. Stable monthly number → flat per-seat (you may overpay for light users). Lower entry plus flexibility → seat-plus-credits (needs model governance or overages surprise you). Pay-only-for-use → pay-per-token API (harder to forecast; spikes happen). There is no universal right answer.
3. Sanity-check free tiers before anyone gets attached. Compare the quota to expected per-developer usage, confirm per-user vs. shared quota, and find out what happens when exceeded — hard stop, throttling, or silent overage.
4. Set model governance before rollout. Define defaults for routine work, approve premium models only for genuinely hard problems (with criteria), and review monthly during adoption, quarterly once things stabilize. Model selection is a budget decision disguised as a settings preference.
5. Plan for scale explicitly. The cheapest option for a 10-person pilot is often not the cheapest at 40 developers. Re-run the evaluation when team size or usage shifts.
6. Align with finance before you commit. Provide projected cost under low, moderate, and high usage scenarios, a plain-language explanation of the billing model and its variability drivers, and a spend review plan — TrueFoundry's teams use a 90-day checkpoint.

The transparency case: why the invoice doesn't match the plan

TrueFoundry's motivating example is the one agencies hear constantly: a team buys a "$20 per seat per month" plan, and six weeks later someone asks why the invoice doesn't match the plan everyone thought they bought. Finance budgets from per-seat estimates; AI coding agent invoices reflect consumption. That mismatch causes approval and reconciliation problems downstream.

Transparency changes behavior: teams that see usage and spend data change how they use agents more than any policy doc does. The practical mitigations are the same as for agency delivery cost — budget the loops, not just the tokens, govern model selection, and insist on budget rails before rollout.

Agent harness rankings (August 2026): Claude Code vs Codex CLI

Claude Code ranks #1 and Codex CLI ranks #2 in the August 2026 agent harness rankings — and the deciding layer is the harness, not the model. Machine Brief's rankings (August 24, 2026) put Claude Code first, Codex CLI second, Cursor third, with Gemini CLI and GitHub Copilot rounding out the top five. explainx.ai's updated top-10 list (published July 22, 2026; updated August 24, 2026) agrees on the top two and fills out the field: Claude Code, Codex CLI/ChatGPT Work, Cursor, Google Antigravity, GitHub Copilot (agent mode), Devin, Windsurf, Factory (Droid), Amazon Q Developer, and Replit Agent.

"Claude Code ranks first on depth of hooks, subagents and dynamic workflows, and is named the default choice for long autonomous coding sessions. Codex CLI leads on cloud, pull-request-shaped autonomy. Cursor holds in-editor workflows, with Gemini CLI and GitHub Copilot rounding out the top five." — Machine Brief, August 24, 2026

Why the top two matter for budgeting: they run the same frontier models. "Several of these tools run the same frontier models. Claude Code and Codex CLI aren't competing over whose underlying brain is smarter. What differs is session management and how gracefully they fail" (Machine Brief, Aug 24, 2026). The same report cites Linear telemetry showing coding agents tripled pull requests without cutting cycle time — "a better harness raises throughput without raising speed" — because the bottleneck sat in review, not generation. The pricing takeaway is direct: you pay for the harness in throughput and for the model in tokens.

The rankings also admit what they don't measure: "They score depth of hooks, subagents and workflow dynamics. They don't measure the things that actually eat a quarter: onboarding friction, cost per session, or what happens to your repo when an agent goes wrong and nobody notices." Treat the ranking as the shortlist — and the cost-per-session math above as the budget line.

Cursor vs Claude Code vs Copilot: what the OpenAI split changes

On August 28, 2026, OpenAI notified SpaceX that it intends to wind down the contract providing OpenAI models to Cursor, with a proposed shutoff date of November 12, 2026 — after SpaceX completed its $60B all-stock acquisition of Cursor on August 14, 2026. OpenAI also said it will not provide future models to Cursor, including its upcoming Astra model, citing an inability to be confident SpaceX will use its technology within OpenAI's terms of service (OpenAI, Aug 28, 2026; Business Insider, Aug 28, 2026; TNW, Aug 29, 2026).

Cursor says OpenAI models served about 5% of its user traffic. Cursor co-founder and CEO Michael Truell responded on August 29, 2026 that "OpenAI models serve about 5% of Cursor user traffic, and we're speaking with the OpenAI team to resolve this" — meaning the other ~95% already flows through Anthropic Claude, Google Gemini, and xAI Grok, including Grok 4.5, which Cursor and SpaceXAI shipped in July 2026 and which is available on every Cursor plan (wccftech, Aug 29, 2026; Cryptobriefing, Aug 29, 2026; Cursor docs).

Anthropic is expanding Claude compute for Cursor, not retreating. Anthropic co-founder and Chief Compute Officer Tom Brown said "we'll continue to increase compute to support Claude models in Cursor," building on the existing Anthropic–SpaceX Colossus 1 compute partnership announced May 6, 2026 (wccftech, Aug 29, 2026; Anthropic, May 6, 2026). For agency budgeting, the practical read: Claude capacity in Cursor is expanding in the near term even as OpenAI models leave.

What the verdict changes for agencies: Cursor remains a strong in-editor harness — but its OpenAI-model workflows have a hard end date. Agencies should audit which Cursor workflows actually depend on OpenAI models (Cursor's own claim is ~5% of traffic, and per-team mixes vary), test Claude Code and Copilot on their own repos before November 12 rather than switching blind, and re-price per-developer model spend now that Claude compute in Cursor is expanding and OpenAI access is shrinking. See the full timeline in OpenAI Is Cutting Off Cursor: What It Means for AI Coding Agents, and run the per-developer cost math on aiagencycalculator.com/cursor-model-cost-shift.

Claude Code vs Codex CLI: strengths, limitations, and best-use cases

Claude Code is "the reference implementation most other harnesses get compared against" (explainx.ai): terminal-first, with /loop for autonomous retry cycles, subagent delegation, hooks for lifecycle events, and CLAUDE.md as a persistent project-memory file the harness reads on every session. Its permission model — prompting before file writes and shell commands unless explicitly trusted — set the default UX pattern most competitors now copy.

• Strengths: the deepest hooks, subagents, and dynamic workflows of any harness; the default choice for long autonomous coding sessions; a permission model that suits supervised agency work.
• Limitations: terminal-native, so less approachable for non-terminal workflows; weekly usage limits apply on Pro/Max/Team plans (see the September 14 limit change above) unless you're on consumption-based Enterprise pricing; heavy autonomy can hit quota walls mid-delivery.
• Best use: long autonomous coding sessions, agent teams that need lifecycle hooks and persistent project memory, and agencies that want auditable permission prompts around client work.

Codex CLI "leads on cloud, pull-request-shaped autonomy" (Machine Brief). It's now folded into a single desktop app alongside Chat and Work modes, sharing usage limits across all three: Work mode leans toward broader "deliverable" tasks — documents, spreadsheets, and reports — while Codex mode stays scoped to repositories (explainx.ai).

• Strengths: cloud-based autonomy shaped around pull requests; Work mode covers deliverables beyond code; a natural fit when your delivery workflow is PR review rather than long-lived terminal sessions.
• Limitations: closed-source; one usage pool shared across Chat, Work, and Codex modes, so heavy multi-mode use burns it faster; cloud-centric by design, which is a poor fit for air-gapped or offline work.
• Best use: teams that ship through pull requests, mixed coding-and-deliverable workflows, and agencies that want agent work to land as reviewable PRs rather than live edits.

Budgeting for a Codex-based workflow? See our Codex CLI pricing guide — Plus, Pro 5x, Pro 20x, Business, and API tiers with usage limits and banked resets.

Which agent harness should agencies standardize on?

Both sources reframe the buying decision the same way: choose the harness layer — context management, subagent coordination, failure handling — before you argue about models. explainx.ai's practical guidance:

• Least setup friction: start with Claude Code or Cursor — "both have the most polished onboarding and sensible permission defaults out of the box."
• Enterprise standardization: Factory's Droid model or GitHub Copilot's SDK "give you a platform to build consistent, scoped agent configurations rather than one general-purpose tool everyone configures differently."
• Open-source / self-host / data residency: OpenCode (breadth of providers) or Pi (modify the loop itself); Aider for git-native reviewable commits.

Decision factors, per explainx.ai: compliance and data residency, team size, model flexibility, customization depth, and budget shape. One naming caveat before you standardize on Google's entry: Machine Brief (Aug 24) still lists Gemini CLI in its top five, while explainx says Google Antigravity "replaced Gemini CLI as Google's primary coding-agent surface." Both readings are current as of late August 2026 — check which surface Google is actually shipping before you bet an agency workflow on it.

Agentic engineering vs vibe coding: what the DHH–Lex debate means for pricing and engagement models

On August 26, 2026, DHH — creator of Ruby on Rails, CTO of 37signals, and founder of the Omarchy Linux distribution — spent five hours on Lex Fridman Podcast #501 arguing that the industry's real divide is no longer human vs AI coding, but vibe coding vs agent-accelerated engineering. For AI coding agents for agencies, the episode is a pricing brief: vibe coding is the cheap tier, agent-accelerated engineering the premium one. DHH's operational definition: vibe coding is telling an agent to build software and not looking at the implementation; agent-accelerated development keeps the human engaged with what the agent produces (0:51:05). We cover the full argument, its pushback, and the "will AI replace programmers" question in Agentic Engineering vs Vibe Coding: How AI Coding Agents Are Rewriting Agency Delivery. This section is about what the distinction does to your quote.

When vibe coding is acceptable — and what to charge

Vibe coding is legitimate for scoped, low-risk work: prototypes, internal tools, marketing pages, throwaway scripts, and greenfield experiments where architectural decay is acceptable. DHH himself keeps vibe-coded projects. It is cheap to deliver — agents implement and nobody reviews deeply — and it should be priced that way: fixed fee, fast turnaround, disposable output. The failure mode is scale: when "vibe" output accumulates on a production system, architecture decays (his Basecamp 5 story — designers shipping PRs that "destroyed the architecture of the system", 0:16:44) and someone pays for the cleanup, usually more than the vibe saved.

Why agentic engineering commands premium rates

The premium is not the agent — it is the layer around it. DHH's explanation of why the 10X–1000X productivity claims only materialize when humans don't slow the loop:

"As soon as you're having human teams work together on something, the bottleneck is rarely implementation. It's human bandwidth and communication. … to get that magical 10X, 100X, in a few rare cases, 1000X productivity boost, you have to interact with the agents directly, and you cannot intermediate that bandwidth with another human because it's simply too slow." — DHH, Lex Fridman Podcast #501 (0:18:44), transcript

Two pricing implications follow. First, projects priced on "agent hours" are pricing the wrong unit: the billable unit is the review, the architecture call, and the sign-off. Second, DHH's stated operating procedure is two-model cross-review — one frontier model does the work, a differently sourced model reviews it — which adds a second token bill but catches one vendor's agent blind spots. On production work, that second-model review is a real cost line agencies should quote explicitly rather than absorb.

Projects vs retainers: how the debate settles the engagement model

DHH's own workflow is the retainer argument: agents drafted or closed PRs on a regular schedule, and he made "the final determination" on a shortlist (2:43:27). Continuous work needs a continuous engagement — a retainer with a defined review capacity — because the agent work never stops between milestones. A project quote assumes work ends at delivery; agentic engineering doesn't, and the pricing should say so.

Refreshing the "end of manual programming" claim

Earlier framing of this debate treated AI coding as simple substitution — manual programming ends, agents take over, agencies price accordingly. DHH's actual position is more precise, and more useful for pricing. He argues the economic case for hand-sweating every line is eroding — not the value of builders:

"The reason why I, for 25 years, was sweating every line of code so judiciously was because I knew the payoff of keeping an architecture coherent and malleable was software that could change and evolve quickly with a small team … That was premised on humans doing the modifications. I think it is an open question to which degree this still matters." — DHH, Lex Fridman Podcast #501 (1:00:35), transcript

And the romanticization of handwritten code is already underway:

"It's already happening, the romanticization of handwritten code. And I have some of it because it was a very romantic era. I'm grateful to have been alive for 20 years of economically valuable handwritten code. That was a good time." — DHH, Lex Fridman Podcast #501 (1:04:57), transcript

What this means for pricing: don't sell "handwritten" as a premium, and don't discount "agent-written" as a commodity. The premium is engineering judgment — what to build, what to let the agent build, what to review, and how to keep the system coherent while agents move fast. One counterintuitive implication from the episode: over-prescription damages agent output. DHH on his own workflow: "more often than not, I have the humility to recognize that the agent knows best" (0:55:30, transcript) — so the premium lies in problem definition and outcome review, not in dictating the implementation route. Clients shouldn't pay for heavy-handed specs; they should pay for someone who knows which details to specify and which to leave to the agent.

Analysis: DHH never says "agency" or "client" in the episode; the engagement-model mapping above is our extrapolation of his adoption advice. All quotes are verbatim from the official Lex Fridman Podcast #501 transcript (Aug 26, 2026), with YouTube deep links.

Frequently asked questions

Sources

• Lex Fridman Podcast #501 — "DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux" (published Aug 26, 2026): official transcript · YouTube · Spotify · episode page
• TrueFoundry — "AI Coding Agent Pricing: How to Choose the Right Plan" (Sreejith Jicks, Aug 6, 2026): truefoundry.com/blog/ai-coding-agent-pricing
• Author page: truefoundry.com/blogs/authors/sreejith-jicks
• MacRumors — "Grok Bot Brings Always-On AI Agents to macOS and iOS" (Juli Clover, Aug 11, 2026): macrumors.com/2026/08/11/grok-bot-macos-ios
• SpaceXAI (x.ai) — "Introducing Grok Bot" (Aug 11, 2026): x.ai/news/introducing-grok-bot
• Slack — "Slack Code: Where Your Team and Agents Build Together" (Aug 20, 2026): slack.com/blog/news/slack-code-channels-for-agents
• Salesforce — "Introducing Slack Code: Agentic Coding for Teams" (Aug 20, 2026): salesforce.com/introducing-slack-code
• The Verge — "Slack is launching collaborative vibe-coding channels" (Aug 20, 2026): theverge.com/tech/982628/slack-code-vibe-coding-channels-launch
• Gizmodo — "Slack Has (of Course) Launched a Vibe Coding Tool" (Aug 20, 2026): gizmodo.com/slack-has-of-course-launched-a-vibe-coding-tool-2000800885
• The Register — "Slack Code taps into collective vibe, puts AI agents into the group chat" (Aug 20, 2026): theregister.com/saas/2026/08/20/slack-code-taps-into-collective-vibe-puts-ai-agents-into-the-group-chat/5290413
• Anthropic via @ClaudeDevs on X — announcement of the Sept 14, 2026 limit change (Aug 29, 2026): x.com/ClaudeDevs/status/2093742321473065266 · clarification x.com/ClaudeDevs/status/2093742322525810912
• BleepingComputer — "Anthropic is cutting Claude Code's current weekly limits by 17%" (Aug 29, 2026): bleepingcomputer.com
• Anthropic Help Center — "Claude Code May–August 2026 weekly limits promotion" (updated Aug 18, 2026; boost extended through August 31, 2026 — Help Center not yet updated to the Sept 14 date as of Aug 30): support.claude.com/en/articles/15910845
• Machine Brief — "Claude Code Just Topped the Agent Harness Rankings – and the Harness, Not Model, Decides" (Riku, Aug 24, 2026): machinebrief.com/news/claude-code-agent-harness-rankings-august-2026
• explainx.ai — "Top 10 Closed-Source and Open-Source Agent Harnesses (2026)" (Yash Thakker, Jul 22, 2026; updated Aug 24, 2026): explainx.ai/blog/top-10-open-closed-source-agent-harnesses-2026
• OpenAI — "Our decision on Cursor following its acquisition by SpaceX" (Aug 28, 2026): openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex
• Business Insider — "OpenAI says it's ending its deal with Cursor because Elon Musk's companies violate contracts" (Lloyd Lee, Aug 28, 2026): businessinsider.com/openai-ends-cursor-contract-elon-musk-spacex-sam-altman-feud-2026-8
• Cryptobriefing — "Cursor founder says OpenAI models account for just 5% of user traffic" (Aug 29, 2026): cryptobriefing.com/cursor-openai-5-percent-traffic
• Wccftech — "Anthropic Pounces As OpenAI Abandons SpaceX's Cursor, Vowing to Increase Claude Compute" (Rohail Saleem, Aug 29, 2026): wccftech.com/anthropic-pounces-as-openai-abandons-spacexs-cursor
• The Next Web — "OpenAI to stop supplying models to Cursor after SpaceX acquisition" (Aug 29, 2026): thenextweb.com/news/openai-ends-cursor-contract-spacex-acquisition
• Anthropic — "Higher usage limits for Claude and a compute deal with SpaceX" (May 6, 2026): anthropic.com/news/higher-limits-spacex
• Cursor Docs — Models & Pricing (fetched Aug 29, 2026): docs.cursor.com/account/pricing
• Cursor — Pricing (fetched Aug 29, 2026): cursor.com/pricing

Note: the source guide's governance section promotes TrueFoundry's own AI Gateway product; its latency/RPS claims are vendor marketing. The billing models, cost variables, and budgeting guidance cited here are independent of that.