Is ChatGPT Plus worth $20/month vs the API?

For casual chat (~50 messages/day) the API costs roughly $3–$8/mo, so the subscription is sub-1× value. For heavy coding or research use, the subscription typically delivers 5–15× of equivalent API spend.

Which AI subscription gives the best value for coders?

At the $20 tier, Claude Pro leads on raw token leverage for agentic coding. At $200, Claude Max 20× and ChatGPT Pro deliver an implied $8k–$14k of API value per month when pushed to their weekly caps (SemiAnalysis).

How often is the data updated?

A GitHub Actions cron job re-runs the snapshot every 6 hours and the page revalidates on every visit, so prices stay within a day of the vendor pages.

cold research · auto-refresh

What does your $200/mo
actually buy in tokens?

SemiAnalysis ran a single Claude Max 20× and a single ChatGPT Pro account against long agentic coding tasks until the weekly caps slammed shut. The implied raw-API value was $8,000 for Claude and $14,000 for OpenAI — per month, per seat. This page compares every consumer subscription on the planet, US to China, against the same yardstick.

providers

consumer plans

api models

refresh

daily

── usage profile ──────────────────────────────

Value leaderboard

Equivalent API cost ÷ subscription price. Higher = subscription gives you more raw tokens than buying API directly.

profile: Heavy coder

filters

showing 35 / 35

provider:

tier:

value:

has weekly cap

plan / model

$/mo range

api equiv $/mo min

value × min

weekly cap contains

#
01	OpenAI	ChatGPT Free gpt-5 (limited)	$0	$65	free	few/day
02	Anthropic	Claude Free claude-haiku-4.5	$0	$185	free	~30 msg/5h
03	Google DeepMind	Gemini Free gemini-2.5-flash	$0	$81	free	daily quota
04	xAI	Grok Free grok-4-fast	$0	$25	free	daily quota
05	Perplexity	Perplexity Free sonar	$0	$65	free	few Pro/day
06	Alibaba Qwen	Qwen Chat qwen3-max	$0	$222	free	free
07	Moonshot (Kimi)	Kimi Chat kimi-k2	$0	$99	free	free
08	GitHub Copilot	GitHub Copilot Pro+ gpt-5 / claude-opus-4.1 / gpt-5-pro	$39	$2775	71.2×	1,500 premium requests / month
09	Windsurf (Codeium)	Windsurf Pro claude-sonnet-4.5 / gpt-5 / gemini-3-pro	$15	$555	37.0×	500 prompt credits / month
10	GitHub Copilot	GitHub Copilot Pro gpt-5 / claude-sonnet-4.5 / haiku-4.5	$10	$325	32.5×	300 premium requests / month
11	Cursor	Cursor Pro gpt-5 / claude-sonnet-4.5 / gemini-3-pro	$20	$555	27.8×	Auto+Composer pool + API pool (credit-based since Jun 2025)
12	Anthropic	Claude Pro claude-sonnet-4.6	$20	$555	27.8×	~45 msg/5h, ~40-80h Sonnet/wk
13	Cognition (Devin)	Devin Pro SWE-1.6 (Cognition in-house) + Adaptive router to Claude/GPT/Gemini	$20	$555	27.8×	Quota-based (ACU credits, ~$20 of agent compute)
14	Google DeepMind	Google AI Pro gemini-3-pro	$20	$420	21.0×	~100 prompts/day Pro
15	OpenAI	ChatGPT Pro gpt-5-pro + o3-pro	$200	$3900	19.5×	effectively unlimited heavy use
16	xAI	SuperGrok grok-4	$30	$555	18.5×	~100 msg/2h
17	Perplexity	Perplexity Pro sonar-pro + frontier	$20	$325	16.3×	300+ Pro searches/day
18	OpenAI	ChatGPT Plus gpt-5	$20	$325	16.3×	~80 msg/3h, ~2400/wk soft
19	Cursor	Cursor Pro+ gpt-5 / claude-sonnet-4.5 / gemini-3-pro	$60	$555	9.3×	~3× Pro fast-request pool
20	Anthropic	Claude Max 5x claude-sonnet-4.6 + opus-4.1	$100	$555	5.5×	5x Pro = ~200-400h Sonnet/wk
21	Moonshot (Kimi)	Kimi Moderato kimi-k2 / kimi-code	$19	$99	5.2×	roughly Claude Pro-class usage @ 1/10 price
22	Anthropic	Claude Max 20x claude-opus-4.1 + sonnet-4.6	$200	$555	2.8×	20x Pro, soft weekly cap
23	Cognition (Devin)	Devin Max SWE-1.6 (Cognition in-house) + Adaptive router to Claude/GPT/Gemini	$200	$555	2.8×	Higher ACU quota + priority
24	Cursor	Cursor Ultra gpt-5 / claude-sonnet-4.5 / gemini-3-pro	$200	$555	2.8×	~20× Pro usage incl. cloud agents
25	xAI	SuperGrok Heavy grok-4-heavy	$300	$555	1.9×	multi-agent, highest cap
26	Google DeepMind	Google AI Ultra gemini-3-pro-deep-think + veo	$250	$420	1.7×	highest quota in consumer market
27	Perplexity	Perplexity Max frontier + Comet browser	$200	$325	1.6×	unlimited Labs, Comet agent
28	Ollama	Ollama Pro gpt-oss-120b / qwen3-coder / deepseek-v3	$20	$24	1.2×	GPU-time based; ~50× free tier
29	Mistral	Le Chat Free mistral-medium	$0	—	—	daily quota
30	DeepSeek	DeepSeek Chat deepseek-v3.2	$0	—	—	free unlimited chat
31	Zhipu (GLM)	GLM Chat glm-4.6	$0	—	—	free
32	Z.ai (Zhipu)	GLM Coding Plan · Lite glm-4.6 / glm-5.2	$3	—	—	~3× Claude Pro usage @ ~1/7 the price
33	Mistral	Le Chat Pro mistral-large + magistral	$15	—	—	generous
34	Z.ai (Zhipu)	GLM Coding Plan · Pro glm-4.6 / glm-5.2	$15	—	—	~3× Claude Max $100 usage @ 1/7 the price
35	Z.ai (Zhipu)	GLM Coding Plan · Max glm-4.6 / glm-5.2	$30	—	—	~3× Claude Max $200 usage @ 1/7 the price

Free plans show free when they deliver any API-equivalent value. Sub-1× means you'd save money by going API-only at this usage level.

── budget zone · $5 – $25 / mo ──────────────────────────────

deep dive · all 12 plans

Every subscription between $5 and $25,
compared cold.

This is the entry-tier band — where casual users actually live. We pulled every consumer plan our crawler tracks inside the $5–$25 window, mapped it to the frontier model it actually serves, and ran the math at three usage intensities.

plans found

price spread

$10–$20

avg price

$17.83

best leverage (coder)

37.0×

plan	$/mo	model	class	cap	S/I/V/A/M/L	casual ×	researcher ×	coder ×
Windsurf Pro	$15	claude-sonnet-4.5 / gpt-5 / gemini-3-pro	mid	500 prompt credits / month	○○○○○○	0.30×	4.60×	37.0×
GitHub Copilot Pro	$10	gpt-5 / claude-sonnet-4.5 / haiku-4.5	mid	300 premium requests / month	○○○○○○	0.26×	4.00×	32.5×
Cursor Pro	$20	gpt-5 / claude-sonnet-4.5 / gemini-3-pro	mid	Auto+Composer pool + API pool (credit-based since Jun 2025)	○○○○○○	0.23×	3.45×	27.8×
Claude Pro	$20	claude-sonnet-4.6	frontier	~45 msg/5h, ~40-80h Sonnet/wk	●○○●●●	0.23×	3.45×	27.8×
Devin Pro	$20	SWE-1.6 (Cognition in-house) + Adaptive router to Claude/GPT/Gemini	mid	Quota-based (ACU credits, ~$20 of agent compute)	○○○○○○	0.23×	3.45×	27.8×
Google AI Pro	$20	gemini-3-pro	frontier	~100 prompts/day Pro	●●●●●●	0.17×	2.60×	21.0×
Perplexity Pro	$20	sonar-pro + frontier	frontier	300+ Pro searches/day	●●○○●○	0.13×	2.00×	16.3×
ChatGPT Plus	$20	gpt-5	frontier	~80 msg/3h, ~2400/wk soft	●●●●●●	0.13×	2.00×	16.3×
Kimi Moderato	$19	kimi-k2 / kimi-code	mid	roughly Claude Pro-class usage @ 1/10 price	○○○○○○	0.04×	0.65×	5.2×
Ollama Pro	$20	gpt-oss-120b / qwen3-coder / deepseek-v3	mid	GPU-time based; ~50× free tier	○○○○○○	0.01×	0.15×	1.2×
Le Chat Pro	$15	mistral-large + magistral	near-frontier	generous	●●○○●○	0.00×	0.00×	0.0×
GLM Coding Plan · Pro	$15	glm-4.6 / glm-5.2	mid	~3× Claude Max $100 usage @ 1/7 the price	○○○○○○	0.00×	0.00×	0.0×

S=Search · I=Image-gen · V=Video-gen · A=Agent/code-tool · M=Multimodal in · L=Long-ctx (≥1M). ● = yes · ○ = no.

Casual chat · 0.5M in

🇺🇸 Windsurf Pro0.30×

🇺🇸 GitHub Copilot Pro0.26×

🇺🇸 Cursor Pro0.23×

🇺🇸 Claude Pro0.23×

🇺🇸 Devin Pro0.23×

🇺🇸 Google AI Pro0.17×

🇺🇸 Perplexity Pro0.13×

🇺🇸 ChatGPT Plus0.13×

🇨🇳 Kimi Moderato0.04×

🇺🇸 Ollama Pro0.01×

🇫🇷 Le Chat Pro0.00×

🇨🇳 GLM Coding Plan · Pro0.00×

Researcher · 8M in

🇺🇸 Windsurf Pro4.6×

🇺🇸 GitHub Copilot Pro4.0×

🇺🇸 Cursor Pro3.5×

🇺🇸 Claude Pro3.5×

🇺🇸 Devin Pro3.5×

🇺🇸 Google AI Pro2.6×

🇺🇸 Perplexity Pro2.0×

🇺🇸 ChatGPT Plus2.0×

🇨🇳 Kimi Moderato0.65×

🇺🇸 Ollama Pro0.15×

🇫🇷 Le Chat Pro0.00×

🇨🇳 GLM Coding Plan · Pro0.00×

Heavy coder · 60M in

🇺🇸 Windsurf Pro37.0×

🇺🇸 GitHub Copilot Pro32.5×

🇺🇸 Cursor Pro27.8×

🇺🇸 Claude Pro27.8×

🇺🇸 Devin Pro27.8×

🇺🇸 Google AI Pro21.0×

🇺🇸 Perplexity Pro16.3×

🇺🇸 ChatGPT Plus16.3×

🇨🇳 Kimi Moderato5.2×

🇺🇸 Ollama Pro1.2×

🇫🇷 Le Chat Pro0.00×

🇨🇳 GLM Coding Plan · Pro0.00×

Windsurf (Codeium)

🇺🇸 Windsurf Pro

claude-sonnet-4.5 / gpt-5 / gemini-3-pro

$15/mo

37.0× coder lev.

api equiv (coder)

$555/mo

region

—

cap

500 prompt credits / month

✓ best for: —

! caveat: —

GitHub Copilot

🇺🇸 GitHub Copilot Pro

gpt-5 / claude-sonnet-4.5 / haiku-4.5

$10/mo

32.5× coder lev.

api equiv (coder)

$325/mo

region

—

cap

300 premium requests / month

✓ best for: —

! caveat: —

Cursor

🇺🇸 Cursor Pro

gpt-5 / claude-sonnet-4.5 / gemini-3-pro

$20/mo

27.8× coder lev.

api equiv (coder)

$555/mo

region

—

cap

Auto+Composer pool + API pool (credit-based since Jun 2025)

✓ best for: —

! caveat: —

Anthropic

🇺🇸 Claude Pro

claude-sonnet-4.6

$20/mo

27.8× coder lev.

api equiv (coder)

$555/mo

region

🇺🇸 US · Anthropic

cap

~45 msg/5h, ~40-80h Sonnet/wk

✓ best for: Coding, long-doc analysis, Claude Code CLI light usage

! caveat: Hidden weekly cap (Aug 2025); no image/video generation

Cognition (Devin)

🇺🇸 Devin Pro

SWE-1.6 (Cognition in-house) + Adaptive router to Claude/GPT/Gemini

$20/mo

27.8× coder lev.

api equiv (coder)

$555/mo

region

—

cap

Quota-based (ACU credits, ~$20 of agent compute)

✓ best for: —

! caveat: —

Google DeepMind

🇺🇸 Google AI Pro

gemini-3-pro

$20/mo

21.0× coder lev.

api equiv (coder)

$420/mo

region

🇺🇸 US · Google

cap

~100 prompts/day Pro

✓ best for: Bundle: 2TB Drive + Veo 3 + Imagen + NotebookLM + Gemini in Workspace

! caveat: Veo/Imagen quotas separate & small; Workspace integration only if you live in Gmail/Docs

Perplexity

🇺🇸 Perplexity Pro

sonar-pro + frontier

$20/mo

16.3× coder lev.

api equiv (coder)

$325/mo

region

🇺🇸 US · Perplexity (routes to OpenAI/Anthropic/Google)

cap

300+ Pro searches/day

✓ best for: Citation-grade web research, fact-checking, Spaces

! caveat: You don't pick the model; routing is opaque & changes weekly

OpenAI

🇺🇸 ChatGPT Plus

gpt-5

$20/mo

16.3× coder lev.

api equiv (coder)

$325/mo

region

🇺🇸 US · OpenAI

cap

~80 msg/3h, ~2400/wk soft

✓ best for: Generalist + image (gpt-image-1) + Sora clips + Custom GPTs

! caveat: Silent downgrade to GPT-5-mini after 3h cap; Sora has separate quota

Moonshot (Kimi)

🇨🇳 Kimi Moderato

kimi-k2 / kimi-code

$19/mo

5.2× coder lev.

api equiv (coder)

$99/mo

region

—

cap

roughly Claude Pro-class usage @ 1/10 price

✓ best for: —

! caveat: —

Ollama

🇺🇸 Ollama Pro

gpt-oss-120b / qwen3-coder / deepseek-v3

$20/mo

1.2× coder lev.

api equiv (coder)

$24/mo

region

—

cap

GPU-time based; ~50× free tier

✓ best for: —

! caveat: —

Mistral

🇫🇷 Le Chat Pro

mistral-large + magistral

$15/mo

0.0× coder lev.

api equiv (coder)

$0/mo

region

🇪🇺 EU (Paris) · GDPR-native

cap

generous

✓ best for: EU residency, French/multilingual, cheapest in zone

! caveat: Mistral Large lags GPT-5/Sonnet on reasoning & coding benchmarks

Z.ai (Zhipu)

🇨🇳 GLM Coding Plan · Pro

glm-4.6 / glm-5.2

$15/mo

0.0× coder lev.

api equiv (coder)

$0/mo

region

—

cap

~3× Claude Max $100 usage @ 1/7 the price

✓ best for: —

! caveat: —

verdict · $5–$25 zone

Cheapest entry

GitHub Copilot Pro

$10/mo

If you only need a frontier-ish chat with EU data residency.

Best for researchers

Windsurf Pro

$15/mo

4.60× leverage at 8M-in profile. Long ctx wins.

Best for coders

Windsurf Pro

$15/mo

37.0× leverage at 60M-in profile — the highest in band.

Bottom line. At this price band every plan except Mistral converges on $20. For casual chatters all five are sub-1× value — you'd literally save money paying the API meter. The subscription only starts earning its keep at the researcher profile (~3× for Claude/Gemini) and goes vertical for heavy coders. If you ship code daily, Claude Pro wins on raw token leverage but loses on multimodal breadth — Google AI Pro is the better generalist bundle (Veo 3, Imagen, 2TB Drive, NotebookLM) for the same $20. ChatGPT Plus is the safest default for non-technical users. Perplexity Pro is a tool, not a model subscription — buy it only if your job is web research with citations. Le Chat Pro is the only EU-native option and $5 cheaper, but you trade real frontier quality for it.

api layer · 24 frontier & open-weight models

What it actually costs per million tokens,
if you skip the subscription.

Same models, no monthly cap, no UI: you pay per token via OpenRouter, the official API, or a Chinese mirror. Below — sortable, filterable, blended against the profile you picked above (currently Heavy coder: 60M in / 25M out per month).

models tracked

cheapest input

$0.15

cheapest output

$0.42

avg in / out

$2.37 / $13.65

longest context

2000k

open-weight

9/24

cheapest frontier

GLM-5.2

$194/mo @ Heavy coder

in $1.4 · out $4.4 / Mtok

most expensive (output)

GPT-5 Pro (API)

$120/Mtok output

8.8× the average

best $/M output

DeepSeek V3.2

$0.42/Mtok output

32× cheaper than average

filters

showing 24 / 24

provider

tier

open-weight only

search model

max $/M input

max $/M output

min context (k tokens)

		tier			cache				region
🇺🇸 ollama	gpt-oss-120b (Ollama Cloud)oss	budget	$0.15	$0.60	—	4.0×	128k	$24	🇺🇸 US
🇺🇸 xai	Grok 4 Fast	budget	$0.20	$0.50	$0.050	2.5×	2000k	$25	🇺🇸 US
🇨🇳 deepseek	DeepSeek V3.2oss	budget	$0.28	$0.42	$0.028	1.5×	128k	$27	🇨🇳 CN
🇨🇳 minimax	MiniMax M2oss	budget	$0.30	$1.20	—	4.0×	200k	$48	🇨🇳 CN
🇺🇸 openai	GPT-5 mini	mid	$0.25	$2.00	$0.025	8.0×	400k	$65	🇺🇸 US
🇺🇸 google	Gemini 2.5 Flash	budget	$0.30	$2.50	$0.075	8.3×	1000k	$81	🇺🇸 US
🇨🇳 deepseek	DeepSeek R1oss	mid	$0.55	$2.19	$0.140	4.0×	128k	$88	🇨🇳 CN
🇨🇳 zhipu	GLM-4.6oss	mid	$0.60	$2.20	$0.110	3.7×	200k	$91	🇨🇳 CN
🇨🇳 moonshot	Kimi K2oss	mid	$0.60	$2.50	$0.150	4.2×	256k	$99	🇨🇳 CN
🇺🇸 anthropic	Claude Haiku 4.5	mid	$1.00	$5.00	$0.100	5.0×	200k	$185	🇺🇸 US
🇨🇳 alibaba	Qwen3 Coder Plusoss	mid	$1.00	$5.00	—	5.0×	1000k	$185	🇨🇳 CN
🇨🇳 zhipu	GLM-5.2oss	frontier	$1.40	$4.40	$0.260	3.1×	200k	$194	🇨🇳 CN
🇨🇳 alibaba	Qwen3-Max	frontier	$1.20	$6.00	—	5.0×	256k	$222	🇨🇳 CN
🇫🇷 mistral	Mistral Large 2oss	mid	$2.00	$6.00	—	3.0×	128k	$270	🇪🇺 EU
🇺🇸 openai	o3	frontier	$2.00	$8.00	$0.500	4.0×	200k	$320	🇺🇸 US
🇺🇸 openai	GPT-5	frontier	$1.25	$10.00	$0.125	8.0×	400k	$325	🇺🇸 US
🇺🇸 openai	GPT-5.1	frontier	$1.25	$10.00	$0.125	8.0×	400k	$325	🇺🇸 US
🇨🇦 cohere	Command A	mid	$2.50	$10.00	—	4.0×	256k	$400	🇨🇦 CA
🇺🇸 google	Gemini 3 Pro	frontier	$2.00	$12.00	$0.500	6.0×	1000k	$420	🇺🇸 US
🇺🇸 google	Gemini 3.1 Pro	frontier	$2.00	$12.00	$0.200	6.0×	1000k	$420	🇺🇸 US
🇺🇸 xai	Grok 4	frontier	$3.00	$15.00	$0.750	5.0×	256k	$555	🇺🇸 US
🇺🇸 anthropic	Claude Sonnet 4.6	frontier	$3.00	$15.00	$0.300	5.0×	1000k	$555	🇺🇸 US
🇺🇸 anthropic	Claude Opus 4.1	frontier	$15.00	$75.00	$1.500	5.0×	200k	$2775	🇺🇸 US
🇺🇸 openai	GPT-5 Pro (API)	frontier	$15.00	$120.00	—	8.0×	400k	$3900	🇺🇸 US

Blended cost = profile in-tokens × $/M in + out-tokens × $/M out. out/in ≥ 8× means the model penalizes long generations — batch or stream carefully. Cache column = price when input is served from prompt-cache (where supported). Open-weight tag = model weights are publicly downloadable, so the same model can be self-hosted or served by 3rd-party APIs.

case study · semianalysis

Anthropic Claude Max 20×

paid

$200

→

api equiv

$8,000

leverage

40×

One seat, Sonnet 4.5 via Claude Code, pushed to the new weekly cap. Implied $/token consumed: ≈ $8k of public API list price.

case study · semianalysis

OpenAI ChatGPT Pro

paid

$200

→

api equiv

$14,000

leverage

70×

One seat, GPT-5-Pro + o3-pro through Codex, long agentic runs. Implied $/token consumed: ≈ $14k of public API list price.

Method

01For each consumer plan we map the model the user actually gets when they hit send. Plus → GPT-5. Max 20× → Sonnet 4.5 with Opus burst.
02We multiply your usage profile (input + output tokens / month) by the public list API price of that model. That is the "what it would cost on the meter" number.
03We divide by the subscription price. Result is the leverage multiple. SemiAnalysis observed 40× and 70× when pushed to the weekly caps.
04A daily cron snapshots the table so historical drift is auditable.

Caveats — read these

!Anthropic added an undocumented weekly cap in Aug 2025. Sonnet hours/week vary by account; SemiAnalysis number is the ceiling, not your typical day.
!OpenAI Pro's quoted $14k assumes long-running Codex agents. Chat-only Pro users will see 5–10× leverage, not 70×.
!Chinese frontier models (DeepSeek V3.2, Kimi K2, GLM-4.6, Qwen3-Max) have no paid consumer tier outside China — their value column is "free" but API is also 5–10× cheaper than US frontier.
!Subscription throughput ≠ API throughput. Subs degrade silently (model swaps, smaller context). API is deterministic.

What does your $200/moactually buy in tokens?

Value leaderboard

Every subscription between $5 and $25,compared cold.

What it actually costs per million tokens,if you skip the subscription.

Method

Caveats — read these

What does your $200/mo
actually buy in tokens?

Every subscription between $5 and $25,
compared cold.

What it actually costs per million tokens,
if you skip the subscription.