Skip to content
Littlewolf Labs News
An orange and blue ratchet screwdriver lying in an open case beside rows of interchangeable bits and sockets.

"2023 Wkrętak z bitami" by Jacek Halicki, CC BY-SA 4.0, via Wikimedia Commons.

Model & Plan Comparisons

Cheaper models you can plug into Claude Code: what LongCat, DeepSeek, GLM, MiniMax and Kimi cost (November 2026)

Five model makers publish setup guides for running their models inside Claude Code. We read each one's pricing page on November 2, 2026 and priced the same sample month, from about $3 to $48 in API fees, plus flat-rate plans from $18 a month.

By Anthony Littlewolf Perez

Claude Code is Anthropic's coding agent, and it normally runs on Claude models through a Claude plan or an Anthropic API key. But several other model makers now publish step-by-step guides for pointing Claude Code at their own models instead, usually by changing two settings: the address it sends requests to and the key it uses. Meituan's LongCat-2.5, which we covered in this morning's digest, is the newest. This guide covers five of them: LongCat, DeepSeek, Z.ai's GLM, MiniMax and Moonshot's Kimi. We read every price on the vendor's own page on November 2, 2026, in US dollars before tax. These are list prices, not a quality test. A cheaper model may need more tries to finish the same task.

One caveat comes first. Anthropic's own gateway docs say Anthropic "doesn't support routing Claude Code to non-Claude models through any gateway." The setups below are documented by the model makers, not by Anthropic, so if something breaks, Anthropic won't help.

Anthropic docs saying the company doesn't support routing Claude Code to non-Claude models through any gateway.
Anthropic's Claude Code gateway docs, read November 2, 2026.

How each one charges

  • LongCat (Meituan) charges pay-as-you-go per million tokens for LongCat-2.5. The prices on its page are labeled a "Discounted Price… (limited-time)," with no end date given. Its Claude Code guide sets every Claude Code model slot to LongCat-2.5.
  • DeepSeek charges per million tokens and has two models, deepseek-flash and deepseek-v4-pro. Off-peak rates are half the peak rates. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday. In Pacific Time right now, that is 5 to 8 PM and 10 PM to 2 AM, Sunday through Thursday nights. Weekends are off-peak in full. Its Anthropic API guide covers Claude Code. That guide also notes that some Claude settings are ignored. For example, the "thinking" budget setting is ignored.
  • Z.ai sells the GLM Coding Plan, a subscription paid in weekly credits, not tokens, and lists Claude Code among its supported tools. Each plan has a 5-hour limit and a weekly limit, per its docs. Off-peak use costs half the credits. Peak is 14:00 to 18:00 Singapore time, Monday to Friday, which is 10 PM to 2 AM Pacific, Sunday through Thursday nights.
  • MiniMax charges per million tokens for M3.1 Flash and M3, and also sells an M Plan subscription that works in Claude Code, per its Claude Code guide. We could not see M Plan prices without signing in. More on that below.
  • Kimi (Moonshot AI) charges per million tokens for its API, which has an Anthropic-compatible format for Claude Code. It says its K2 models are closed to new users since October 31 and will be retired by the end of December 2026, which we covered on November 1. It also sells Kimi Code plans for its own coding app.

Pay-as-you-go prices

Prices per million tokens. "Cached input" is input the provider has seen recently, like the same project files sent again, which coding agents do constantly.

ModelInputCached inputOutput
MiniMax-M3.1-Flash$0.15$0.015$0.60
DeepSeek deepseek-flash, off-peak$0.15$0.003$0.60
DeepSeek deepseek-flash, peak$0.30$0.006$1.20
LongCat-2.5 (limited-time price)$0.30$0.006$1.20
MiniMax-M3, up to 512K input$0.30$0.06$1.20
DeepSeek deepseek-v4-pro, off-peak$0.66$0.022$1.98
DeepSeek deepseek-v4-pro, peak$1.32$0.044$3.96
Kimi kimi-k2.7-code (retiring)$0.95$0.19$4.00
Kimi kimi-k3.1-pro$2.00$0.20$10.00
DeepSeek pricing rows for off-peak and peak token rates, with the peak-hours footnote.
DeepSeek pricing, read November 2, 2026.

MiniMax lists M3 at a "Permanent 50% off," so the prices above are after that discount. Kimi's K3 models also charge separately for writing to the cache, at $2 per million tokens for a 5-minute cache on K3.1 Pro, which the table and the sample month leave out.

What a sample month costs

To compare, we priced one month of moderate use: 50 million input tokens, 40 million of them cached, plus 2 million output tokens. Coding agents send far more than they write, and heavy users can run ten times this. Your mix will differ, so treat these as a rough guide.

ModelSample month
DeepSeek deepseek-flash, all off-peakabout $2.82
MiniMax-M3.1-Flashabout $3.30
LongCat-2.5about $5.64
DeepSeek deepseek-flash, all peakabout $5.64
MiniMax-M3about $7.80
DeepSeek deepseek-v4-pro, off-peak / peakabout $11.44 / $22.88
Kimi kimi-k2.7-codeabout $25.10
Kimi kimi-k3.1-proabout $48, plus cache writes

Who it suits. Pay-as-you-go suits people whose use goes up and down, and anyone who wants to try a model before committing. There's no weekly cap to hit, but there's also no ceiling on the bill, so set any spending limit the provider offers.

Flat-rate plans

PlanPriceAllowanceRoughly buys
Z.ai GLM Coding Plan Lite$18 a month ($12.60 a month billed yearly)10,000 credits a week; 2,000 per 5 hoursOur sample month on GLM-5.3 uses about 18,500 credits, or about 4,300 a week, so it fits, though a busy afternoon could hit the 5-hour limit
Z.ai GLM Coding Plan Pro$80 a month ($56 billed yearly)60,000 credits a week; 12,000 per 5 hours6x Lite
Z.ai GLM Coding Plan Max$168 a month ($117.60 billed yearly)140,000 credits a week; 28,000 per 5 hours14x Lite
Kimi Code Plus / Pro / Max / Ultra$19 / $39 / $99 / $199 a month ($180 / $372 / $948 / $1,908 a year)Not published as numbersSee "What we could not verify"
MiniMax M Plan Go / Explore / BuildNot visible without signing in1x / 3x / 7.5x of GoSee "What we could not verify"
Z.ai GLM Coding Plan cards for Lite, Pro and Max, showing the yearly price beside the monthly price.
Z.ai GLM Coding Plan, read November 2, 2026.

Z.ai works out credits from tokens. On GLM-5.3, each 10,000 input tokens costs 6.9 credits, cached input 1.7 and output 24. The cheaper GLM-5.3-Flash costs about a third as much. Z.ai's own table estimates the Lite plan at about 48 to 97 million GLM-5.3 tokens a week at a 95% cache hit rate. Z.ai also offers 20% off quarterly billing.

Who it suits. A flat plan suits steady daily use and people who want a fixed bill. The trade-off is the same one Claude and Grok users know: weekly and 5-hour limits that can stop you mid-task.

For comparison: Claude's own plans

Claude's pricing page lists Pro at $20 a month ($17 a month billed yearly), Max at $100 and Ultra at $200, all with Claude Code included. Anthropic doesn't publish its limits as token counts either. It describes Max as "5x Pro usage" and Ultra as "10x Pro usage every week," with no 5-hour limit for verified Ultra accounts, which we covered on October 30. So a direct dollar-for-dollar comparison with the plans above isn't possible.

How to price your own month

  1. Check how many tokens you actually use. Most providers' consoles show usage by day, split into input, cached input and output.
  2. Split your input into cached and uncached. Coding agents resend a lot of the same context, so a big share of input is often cached, and cached input is 5 to 50 times cheaper on the models above.
  3. Multiply each part by the table prices, and halve DeepSeek's rates for off-peak hours.
  4. For the GLM plan, use Z.ai's formula: (input × 6.9 + cached input × 1.7 + output × 24) ÷ 10,000 credits on GLM-5.3, then compare that with the weekly allowance, not the monthly one.
  5. Try one small real task first. A model that is half the price but needs three tries costs more.

What we could not verify

  • MiniMax M Plan prices. MiniMax's M Plan page lists the Go, Explore and Build tiers and their relative usage, but the public pages we read don't show prices and the subscribe link we tried returned "Page not found" without signing in, so we have no prices to list.
  • Kimi Code allowances, and whether Kimi Code plans work in Claude Code. Kimi's page promises "regularly refreshed usage quotas" without numbers. Its API works in Claude Code, but we found no Kimi page saying the Kimi Code subscription does. Kimi's page also advertises a limited-time 50% off on K2.8 Preview credit use for members.
  • How long discounts last. LongCat's prices are labeled limited-time with no end date.
  • Quality and speed. We didn't test any model. Some Claude Code features, like how much it "thinks," may behave differently or be ignored, as DeepSeek's guide notes.

What to weigh

  • The cheapest rate depends on the hour. DeepSeek's and Z.ai's peak hours fall on weekday evenings and nights in Pacific Time. Daytime work on the West Coast is off-peak for both.
  • Caching matters more than the headline price. On every model here, cached input costs a small fraction of regular input.
  • Watch the discounts. LongCat's prices are temporary, MiniMax calls its M3 discount permanent, and DeepSeek says its prices may change.
  • Kimi's K2 models are going away. If you set one up now, you'll have to move by the end of December.
  • You're on your own for support. Anthropic says it doesn't support Claude Code on other models. The model maker's docs are your help desk.

Sources

Affiliate disclosure: some stories use affiliate or referral links. When a link pays a commission, the sentence next to it says so. Sponsored slots are labeled Sponsored. The channel card is a house ad until a sponsor is booked.

Newsletter

Newsletter coming soon.

Join the waitlist and we'll email you when the first issue launches.

By submitting you agree to our Privacy Policy.

You're on the list after you confirm. We'll email you when the newsletter launches. You can leave from any message. The form posts to Resend.