Deal intelligence for AI
The best AI deals right now
We research pricing, subscriptions, free tiers and promotions and tell you which ones are actually worth using. Updated August 23, 2026.
Today's Deal Spotlight
Kimi K2.7 Code on OpenCode Go: the 10.5x deal hiding behind a scary sticker price
Kimi K2.7 Code lists at $0.95/M input and $4/M output — numbers that make price-comparison sites rank it as expensive. That framing misses how OpenCode Go actually works. Your $10 buys a $60 usage pool, and Kimi's request cap (about 6,750/month) means you can only draw ~$105 of list-price value through it — a 10.5x effective multiple, the highest of any route we track. The catch: each agent-style request costs about $0.016 at list rates (it reads a lot of cached context), so heavy users hit the cap; light users never will and should just pay as-you-go.
- Effective multiple 10.5x — computed, not marketing
- ~6,750 coding-optimized requests/month included
- Vision + tools + JSON support
Fresh from the discovery engine
hermes-discovered · FactJudge-verified · pending editorial review
ChatGPT Plus ($20/mo, Codex included) + OpenAI API GPT-5.6 family
VERBATIM_ONLY
ChatGPT Plus at $20/month includes Codex on web, CLI, IDE extension, and iOS plus the GPT-5.6 family (Sol/Terra/Luna) and ChatGPT credits; p
FactJudge — · checked 2026-08-26 · machine-discovered, editorial review pending
Claude Pro / Max (Claude Code included)
VERBATIM_ONLY
Pro ($20/mo, or $17/mo-equivalent via $200/yr prepaid) includes Claude Code with at least 5x Free usage per rolling 5-hour session plus week
FactJudge — · checked 2026-08-26 · machine-discovered, editorial review pending
Cursor
SUPPORTED
Hobby tier is free with no credit card and includes limited Agent requests plus access to Composer (numeric request cap not published on off
FactJudge 1.00 · checked 2026-08-25 · machine-discovered, editorial review pending
DeepSeek API (pay-as-you-go, off-peak discount)
SUPPORTED
Pure pay-as-you-go token billing (no monthly fee or free tier): deepseek-v4-flash $0.22/$0.66 per 1M in/out peak ($0.007 cache-hit input), d
FactJudge 0.80 · checked 2026-08-25 · machine-discovered, editorial review pending
GitHub Models (free tier)
VERBATIM_ONLY
NONE - GitHub Models was fully retired as of July 30, 2026; playground, model catalog, inference API and BYOK are unavailable to any custome
FactJudge — · checked 2026-08-26 · machine-discovered, editorial review pending
Kimi Code (Moderato plan) + Kimi K3 API
SUPPORTED
Kimi membership subscription (Moderato $19/mo) includes Kimi Code benefits in CLI, VS Code, and third-party tools sharing one quota that ref
FactJudge 1.00 · checked 2026-08-26 · machine-discovered, editorial review pending
NVIDIA NIM / API Catalog free developer tier (hosted inference at integrate.api.nvidia.com)
WEAK
Free trial tier of NVIDIA-hosted NIM inference endpoints (OpenAI-compatible base URL https://integrate.api.nvidia.com/v1): sign in with an N
FactJudge 0.39 · checked 2026-08-26 · machine-discovered, editorial review pending
OpenCode Go
VERBATIM_ONLY
$10/mo buys $12 of usage per 5 hours, $30 weekly and $60 monthly (dollar-defined limits) across ~25 curated open coding models; per-model es
FactJudge — · checked 2026-08-26 · machine-discovered, editorial review pending
OpenCode Zen free models
WEAK
$0 per token (input, output, cached read) on all seven free models listed in the Zen pricing table, served through Zen's pay-as-you-go accou
FactJudge 0.12 · checked 2026-08-26 · machine-discovered, editorial review pending
DeepSeek API — V4 peak/off-peak repricing
SUPPORTED
Pure pay-as-you-go token billing under the new peak/off-peak scheme effective 2026-08-16 16:00 UTC: deepseek-v4-flash $0.44/$1.32 per 1M in/
FactJudge 0.59 · checked 2026-08-26 · machine-discovered, editorial review pending
Perplexity Sonar API (paid tiers)
WEAK
Pay-as-you-go via prepaid credits (no monthly plan, no free tier; Tier 0 accounts start at $0 purchased): sonar $1/$1 per 1M in/out + $5-$12
FactJudge 0.43 · checked 2026-08-25 · machine-discovered, editorial review pending
Qwen Code via Alibaba Cloud ModelStudio Coding Plan Pro + Qwen3.7-Max API (pay-as-you-go)
VERBATIM_ONLY
Coding Plan Pro ($50/mo, fixed fee, for AI coding tools incl. Qwen Code): 6,000 requests per 5 hours + 45,000 per week + 90,000 per month ac
FactJudge — · checked 2026-08-26 · machine-discovered, editorial review pending
Devin Desktop (renamed from Windsurf Jun 2, 2026) - Devin subscription plans
WEAK
Free plan: limited Devin usage and DeepWiki/Review access, but only SWE-1.6 (free models cost 0 tokens); Pro $20/mo (legacy $15/mo subscribe
FactJudge 0.11 · checked 2026-08-26 · machine-discovered, editorial review pending
Windsurf (Devin Desktop) IDE - Free and Pro plans
SUPPORTED
Free ($0): light daily+weekly token-budget quota, limited model availability, unlimited inline edits and unlimited Tab completions; Pro ($20
FactJudge 0.94 · checked 2026-08-25 · machine-discovered, editorial review pending
GLM Coding Plan — Lite
SUPPORTED
Lite: 2,000 credits per 5-hour cycle plus 10,000 credits per week (~43-105M GLM-5.3 tokens/week depending on cache hit rate and peak/off-pea
FactJudge 1.00 · checked 2026-08-25 · machine-discovered, editorial review pending
Rising in search right now
radar + budget →
“deepseek v4 release” · Deepseek
demand 12.05 · monitor
“cursor free alternative” · Cursor
demand 3.06 · monitor
“cursor pricing” · Cursor
demand 2.95 · monitor
“is cursor down” · Cursor
demand 2.48 · monitor
Google autocomplete + trending-now demand signals via SerpApi — they decide where the free crawler looks next.
Best current deals
View all deals#1
OpenCode Go + MiMo-V2.5
Best overall deal right nowWhat you get
Pay $10/month, get $60 of API usage. That is $60 worth of MiMo-V2.5 coding model calls for every $10 you spend.
Real numbers
MiMo-V2.5 gets you ~150,000 typical requests per month. First month is only $5.
How to get it
Sign up at opencode.ai/go, subscribe to Go, start using MiMo-V2.5.
Why it is good
MiMo is a genuinely strong coding model. The usage allowance is 6x what you pay - unusual for any AI subscription.
What to watch out for
OpenCode can change limits or model lineup at any time.
Best for
People who run coding agents or write a lot of code with AI.
#2
OpenCode Go + HY3
Best for agent orchestrationWhat you get
Same $10/month plan, but HY3 is a cheaper-to-run model. You get ~21,500 requests/month of orchestration and planning.
Real numbers
HY3 costs $0.10/M input vs MiMo at $0.14/M, so your $60 allowance stretches further.
How to get it
Same Go subscription. Select HY3 as your model.
Why it is good
Best cheap orchestration model available. Great for routing tasks between other models.
What to watch out for
HY3 is a specialist - not good for general coding.
Best for
Multi-agent systems, task routing, planning.
#3
Gemini API Free
Best completely free optionWhat you get
Use Google Gemini models for free. Input and output are both $0. Access to models with up to 1M token context windows.
Real numbers
No monthly fee. Free within your project rate limits. If you exceed, you pay standard rates.
How to get it
Create a Google AI Studio account at ai.google.dev. Start using the API immediately.
Why it is good
Huge context windows (1M tokens) at zero cost. Hard to beat for experimentation.
What to watch out for
Rate limits vary by model and project. Not guaranteed for production.
Best for
Learning, prototyping, experimenting with large contexts.
#4
GroqCloud Free
Best speed at zero costWhat you get
Use Groq extremely fast inference hardware for free. Each model has its own rate limits (RPM, RPD, TPM).
Real numbers
GPT-OSS 120B: 30 requests/min, 1,000/day. Llama 3.3 70B: 30 RPM, 1,000 RPD.
How to get it
Sign up at console.groq.com. No credit card needed.
Why it is good
Fastest inference available. Good for testing or low-volume production.
What to watch out for
Rate limits are tight. Not for heavy usage.
Best for
Speed-critical prototyping, model testing.
#5
OpenRouter Free + Funded
Best multi-model accessWhat you get
50 free requests/day on 25+ models. Add $10 (spendable credit, does not expire) to unlock 1,000 free-model requests/day.
Real numbers
Free tier: 50 RPD. Funded tier: 1,000 RPD. Credit remains usable for paid models too.
How to get it
Sign up at openrouter.ai. To unlock: add $10+ credit balance.
Why it is good
Access many models through one API. The $10 unlock is spendable, not a fee.
What to watch out for
Free-model requests only. Paid models cost extra on top of your balance.
Best for
Comparing models, flexible routing, low-volume multi-model use.
All OpenCode Go model routes
38 OpenCode Go routesScore = overall value ranking (quality x volume x economics). Effective cost = what you actually pay per request after the $60 usage pool is applied.
| Score | Model | Context | List $/M In | List $/M Out | Eff $/M In | Eff $/M Out | Req/Mo | Eff. Value | Badges | Role |
|---|