Morph Ultra Review: $200 Unlimited Kimi and DeepSeek, With Limits

Morph, a small Y Combinator-backed company that hosts open-weight models for coding agents, launched a flat-rate plan on November 9, 2026. Morph Ultra costs $200 a month and drops the per-token meter on two models: Kimi K3.1 and DeepSeek V4.1 Flash. The plan is capped at 1,000 subscriptions, and the launch post framed it as unlimited coding for a flat fee.

The word unlimited does a lot of work in that pitch. The plan has no monthly token cap, but every subscription carries per-minute rate limits on both models, and Morph reserves no capacity, so a request can fail with a 429 error when the service is busy. Within a day of launch, early subscribers were already reporting slowdowns, and Morph had raised the limits threefold. This article works through what the $200 buys, where the limits bite, and the usage level at which the plan pays for itself.

What Morph Ultra actually is

Morph is the trading name of AutoInfra, Inc. It was part of Y Combinator's Summer 2023 batch, and its Y Combinator page lists a team of three. The sole active founder is Tejas Bhakta, who previously worked on inference at Tesla Autopilot. The company launched on Hacker News in July 2025 with Fast Apply, a model that merges code edits written by larger models into existing files, and now sells hosted open-weight models alongside its own editing, search, and context-compaction models.

Ultra is Morph's second subscription. The existing Scale plan costs $100 a month and comes with 20 million credits, which Morph values at $200. Ultra costs twice that and covers three model IDs with no token meter: morph-kimik31 (Kimi K3.1), morph-dsv41flash (DeepSeek V4.1 Flash), and morph-dsv41flash-1017, the October 17 snapshot of the same model. Every other model on Morph, including the faster morph-kimik3-fast and the restricted morph-dsv41flash-abliterated, bills per token at list price. The plan includes no credits, so other-model usage draws on prepaid credits first, then bills to the card.

One detail changed within a day of launch. Sulat's launch-day write-up listed a fourth included model, morph-kimik3 (plain Kimi K3), and said the two Kimi models shared one pool of limits. The current documentation lists only three IDs, and Kimi K3 is not among them. Morph appears to have narrowed the plan to Kimi K3.1. There is a second naming wrinkle, reported by Sulat: Morph's pricing page labels morph-kimik31 "Kimi K3.1 Pro", while the Ultra documentation calls the same model "Kimi K3.1". The two names describe the same model ID.

Setup is a copy-paste prompt for Claude Code, Codex, Pi, OpenCode, T3 Code, OMP, and Hermes Agent. For Claude Code by hand, the docs point Anthropic's environment variables at Morph's server, with morph-kimik31 as the main model and morph-dsv41flash as the smaller, faster one. Morph serves the OpenAI and Anthropic wire formats, so the same key also works with other agents that accept a custom base URL.

What it costs

The plan costs $200 a month. At November 10, 2026 market rates (1 USD = 0.7454 GBP, 1.4103 CAD), that is about £149.08 or CA$282.06. Morph bills in US dollars, so treat the conversions as approximations.

An organization subscription is billed per member and covers the organization's API keys, but all members share one set of rate limits. Ultra replaces any existing Morph subscription: the unused part of the old plan comes off the first charge, and remaining subscription credits do not carry over. When an Ultra subscription ends, use of the same models goes back to per-token billing.

The terms of service, last updated October 23, 2026, say payments are non-refundable except as required by law or for SLA breaches, and prices can change with 30 days' written notice. On data, Morph keeps prompts and outputs only to run the service and show customers their request history, deletes them within 30 days, and says it does not use customer inputs or outputs to train, fine-tune, or evaluate its models. Full zero data retention is limited to enterprise customers under a signed addendum.

The limits behind "unlimited"

Two meters apply per model, both counted per minute. One tracks uncached input, meaning fresh text the server has not already processed and cached. The other tracks output, what the model writes back. Cached input does not count toward either limit.

The Kimi output cap is the binding constraint: 6,162 tokens a minute is about 103 tokens a second, roughly one model writing flat out. Morph's own product menu lists Kimi K3 at 100 tokens a second, so three Kimi agents sharing one subscription would get about a third of that each. The input cap matters for coding agents that send large files: a fresh 100,000-token prompt uses a little over a minute of the Kimi input allowance. A request bigger than a limit still runs, and the next request waits until the limit has caught up. Output counts when a response finishes. The 429 response states the limit reached, and Retry-After says how long to wait.

Then there is capacity. The plan does not reserve capacity. When Morph is at capacity, requests can return a 429 error until capacity frees up, whether or not a subscriber has reached their own limits. The documentation tells developers to retry with exponential backoff and honor Retry-After when present, and points customers who need reserved capacity and a service-level agreement to Morph's paid dedicated endpoints.

One more thing to note: the limits above are the current figures, read from Morph's documentation on November 10. On launch day they were a third as high for uncached input: 24,290 tokens a minute for Kimi and 701,079 for DeepSeek, as recorded in Sulat's launch-day write-up. Morph raised them after early subscribers reported throttling.

The break-even math

Without the plan, Morph's own pay-as-you-go rates are $2.50 per million input tokens and $14.00 per million output tokens for Kimi K3.1, with cached input at $0.29, and $0.12 input, $0.008 cached, and $0.80 output for DeepSeek V4.1 Flash. Against those rates, the $200 fee breaks even at about 14.3 million Kimi output tokens, or 250 million DeepSeek output tokens. At the per-minute caps, that is roughly 39 hours of continuous Kimi output, or about 54 hours of DeepSeek output.

Note the asymmetry: Kimi output costs 17.5 times as much as DeepSeek output per token, so the plan pays for itself far faster on Kimi-heavy work. A subscriber who mostly runs DeepSeek needs an enormous amount of output before the flat fee beats the meter.

Then there is Morph's own value claim. An engineer at Morph, Nick Khami, posted a company graphic headed "You pay for 1 dot. You get up to 210." Each dot stands for $200 of tokens at list prices: 70 times the fee at 70% cache, 74 times at 80%, 88 times at 90%, 102 times at 95%, 117 times at 97%, and 210 times, or $42,000, at 99% cache. Sulat rebuilt the sum from the launch-day limits and list prices and landed on Morph's figures at 70, 80, and 90% cache, but got about 334 times the fee at 99% cache, higher than Morph's chart. Morph's graphic understates its own theoretical maximum, which shows how sensitive the number is to the assumptions behind it.

My own check, at the current limits, puts the ceiling at about $25,146 of uncached input plus output over 30 days if both model pools ran at their caps around the clock, or about 126 times the fee. Cached input at 99% would add roughly $162,000 more on paper. All of that assumes a month of nonstop traffic at the rate caps with Morph never turning a request away. It is a measure of list-price value, not money saved, and it only applies to someone who would otherwise have paid Morph's per-token rates for that volume.

One more comparison matters. Morph's Kimi list price is above Moonshot's own: Moonshot lists Kimi K3.1 Pro at $2 per million input tokens and $10 output, while Morph charges $2.50 and $14.00, or 25% more on input and 40% more on output. On DeepSeek the comparison flips on input: DeepSeek's own rate card, as covered in our DeepSeek V4.1 Flash review, is $0.15 input and $0.60 output off-peak, so Morph is cheaper on input at $0.12 and pricier on output at $0.80. Anyone weighing Ultra against buying direct should use the vendor's own price, not Morph's, as the baseline.

The launch-day reality

Early reports were mixed. One subscriber, posting as PandaSt0rm, said Kimi K3.1 ran at "between 40 and 70" tokens per second, then reported that with "barely" a 70% cache hit rate, Kimi was "restricted to working in super short 'bursts' for like 10s before its rate limited", and that DeepSeek "seems to be a lot better". The same user later posted a screenshot captioned "k31, 13 minutes for a single turn". Another developer, Hal Wayland, called the Kimi limits "completely unusable".

Khami replied to PandaSt0rm promising changes: higher limits for new sessions and vision support. The documentation, updated November 9 at 23:55 UTC, now shows uncached-input limits three times the launch-day figures. Morph's own product page lists median speeds as of November 9, 23:00 UTC: Kimi K3.1 at 129 tokens a second with a 1.67-second first token, and DeepSeek V4.1 Flash at 156 tokens a second with a 0.74-second first token. Those are Morph's own measurements, not independent tests, and they postdate the worst launch-day reports.

Asked why only 1,000 subscriptions, founder Tejas Bhakta answered "gpus dont grow on trees". Asked about responsiveness, he said there were "no guarantees but just as fast our main traffic". The plan is a capacity bet as much as a pricing bet: Morph is selling a fixed number of seats on hardware it does not promise to reserve for any one of them.

How it compares

The closest flat-rate rival on price is Perplexity Teammates at $200 a month, tied to a Max subscription, but that is a different product: persistent AI teammates, not a model API. Inside Morph's own lineup, the choice is between Ultra and Scale.

Scale is the better deal for anyone whose usage is mixed across models, since its credits work everywhere and carry practically no rate limits. Ultra only wins for heavy, sustained use of exactly two models, and mostly on the Kimi side. A developer running Claude Code against Morph should also remember that the setup routes the main model through morph-kimik31: the Claude Code plan underneath still matters, and Morph's plan does not replace it. Our Kimi K3.1 Pro review and our roundup of the 7 best AI coding agents in 2026 cover the models and the agents this plan is built for.

Common mistakes

First, reading "unlimited" as unthrottled. The plan has no monthly token cap, but the per-minute limits and the lack of reserved capacity are the real constraints, and launch-day users hit both.

Second, buying for the wrong Kimi. The plan covers Kimi K3.1 under the model ID morph-kimik31. Plain Kimi K3, under morph-kimik3, appeared in launch-day coverage but does not appear in the current documentation, and the faster morph-kimik3-fast bills per token.

Third, sharing one organization subscription across a team and expecting per-member limits. The documentation says an organization subscription pools all members under one set of rate limits, so a team shares the same 6,162 Kimi output tokens a minute.

Fourth, treating the $42,000 figure as savings. It is the list price of a month of nonstop traffic at the rate caps, at 99% cache. A subscriber who would not have bought that volume at list price saves nothing like that amount.

Fifth, assuming credits carry over. Switching to Ultra replaces the existing subscription, and remaining subscription credits do not carry over, though the unused part of the old plan comes off the first charge.

Is Morph Ultra really unlimited?

No. There is no monthly token cap on the two included models, but per-minute rate limits cap how fast you can use them, and Morph reserves no capacity, so requests can return a 429 error when its servers are busy. Cached input does not count toward the limits, which is why heavy caching is the whole game on this plan.

How many tokens does $200 buy on pay-as-you-go?

At Morph's list prices, $200 buys about 14.3 million Kimi K3.1 output tokens, or 250 million DeepSeek V4.1 Flash output tokens. On input, the same $200 buys 80 million Kimi tokens or about 1.67 billion DeepSeek tokens. Cached Kimi input is cheaper still, at $0.29 per million.

Can I use Morph Ultra with Claude Code?

Yes. Morph provides a copy-paste setup prompt for Claude Code, Codex, Pi, OpenCode, T3 Code, OMP, and Hermes Agent. For Claude Code by hand, set ANTHROPIC_BASE_URL to https://api.morphllm.com, ANTHROPIC_AUTH_TOKEN to your Morph key, ANTHROPIC_MODEL to morph-kimik31, and ANTHROPIC_SMALL_FAST_MODEL to morph-dsv41flash.

What happens when I hit a rate limit?

A request bigger than a limit still runs, and the next request waits until the limit has caught up. The 429 response states the limit you reached, and Retry-After says how long to wait. The OpenAI and Anthropic SDKs retry with backoff automatically.

Is there a free trial?

Morph lists no trial for Ultra. Pay-as-you-go starts with a first top-up from $10 plus $5 free, and the pricing page lists 200 requests free every month. Morph caps the plan at 1,000 subscriptions.

What happens to my Scale credits if I switch?

They do not carry over. Ultra replaces the existing subscription, the unused part of the old plan comes off the first charge, and any remaining subscription credits are lost. When an Ultra subscription ends, use of the same models goes back to per-token billing.

Morph Ultra is a good deal for one narrow profile: a developer running Kimi K3.1 through a coding agent for hours a day, with a prompt cache doing most of the work. For anyone else, the per-minute limits, the shared capacity, and the two-model scope make the $200 harder to justify than the launch post suggests. The plan is worth a month of real usage before renewing, not a year of faith. This is general information, not financial or purchasing advice.

Sources

• Morph Ultra product page — $200 a month, 1,000 subscriptions, rate limits, pay-as-you-go rates, median speeds, setup instructions (read November 10, 2026)
• Morph Ultra plan documentation — included model IDs, rate limits, capacity terms, organization billing, cancellation (updated November 9, 2026, 23:55 UTC)
• Morph pricing page — pay-as-you-go, Scale $100 a month with 20 million credits, Ultra $200 a month (read November 10, 2026)
• Sulat, "Morph Ultra: unmetered Kimi K3 and DeepSeek for $200 a month" — launch-day limits, $42,000 graphic rebuild, subscriber reports, founder replies, terms of service (November 10, 2026)
• Moonshot AI platform pricing — Kimi K3.1 Pro $2 input, $10 output per million tokens (verified November 10, 2026)
• AI Core Updates, "DeepSeek V4.1 Flash 1017: The $0.15 Coding Model and Its New Search API" — DeepSeek official rate card $0.15/$0.60 off-peak, double at peak (October 23, 2026)
• RateMorph USD to GBP, November 10, 2026 — 1 USD = 0.7454 GBP
• RateMorph USD to CAD, November 10, 2026 — 1 USD = 1.4103 CAD (November 10, 2026)