cheapos
Changes

Find the model.
Skip the bill.

Prices for 436 AI models, free tiers decoded, and reliability measured on real agent work.

Free with tool callingLongest free contextUnder $0.10 / MBest free for codingFree tiers by provider
Models priced
436
refreshed hourly
Free right now
20
see them
Routes ending
24
next: Oct 8
Tokens measured
250.8M
signed cheapoS reports
Most reliable free · measuredNVIDIA: Nemotron 3 Super (free)93.9% answered, likely 89–97% · 164 outcomesTop free coding score · benchmarkQwen: Qwen3.8 27B (free)Coding index 68.1 · Artificial AnalysisLongest free context with tools · publishedThinking Machines: Inkling (free)1,048,576 tokens of context at $0

Browse models

20 models · Free

ModelPrice / MContextCodingReliability measuredAvg time
NVIDIA: Nemotron 3 Super (free)nvidia/nemotron-3-super-120b-a12b:freetools · reasoningFree262K37.7
93.9%likely 89–97% · 164 outcomes
13.2 s
Dots Studio: Dots3-Note Preview (free)dots-studio/dots-3-note-preview:freetools · vision · reasoningFree512K—
92.3%likely 83–97% · 65 outcomes
17.0 s
NVIDIA: Nemotron 3 Ultra (free)nvidia/nemotron-3-ultra-550b-a55b:freetools · reasoningFree1M49.3
89.2%likely 81–94% · 93 outcomes
26.8 s
Poolside: Laguna XS 2.1 (free)poolside/laguna-xs-2.1:freetools · reasoningFree262K—
68.4%likely 53–81% · 38 outcomes
16.3 s
Poolside: Laguna S 2.1 (free)poolside/laguna-s-2.1:freetools · reasoningFree262K—
72.7%likely 52–87% · 22 outcomes
40.1 s
inclusionAI: Ling 3.0 Flash Sante (free)inclusionai/ling-3.0-flash-sante:freetools · reasoningFree262K—
58.3%likely 39–76% · 24 outcomes
11.2 s
NVIDIA: Nemotron 3 Nano Omni (free)nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:freetools · vision · reasoningFree256K13.8
53.8%likely 36–71% · 26 outcomes
19.6 s
Google: Gemma 4 31B (free)google/gemma-4-31b-it:freetools · vision · reasoningFree262K43.4
5%likely 1–24% · 20 outcomes
6.8 s
Google: Gemma 4 26B A4B (free)google/gemma-4-26b-a4b-it:freetools · vision · reasoningFree262K39.3
0%likely 0–13% · 26 outcomes
7.9 s
Cohere: North Mini Code (free)cohere/north-mini-code:freetools · reasoningFree256K36.5
Too few13 of 17 answered · rates start at 20
11.2 s
Free Models Routeropenrouter/freetools · vision · reasoningFree200K—No reports—
Google: Lyria 3 Clip Previewgoogle/lyria-3-clip-previewvisionFree1.05M—No reports—
Google: Lyria 3 Pro Previewgoogle/lyria-3-pro-previewvisionFree1.05M—No reports—
LiquidAI: LFM2.5-2.6B (free)liquid/lfm-2.5-2.6b:freetools · reasoningFree66K—
Too few9 of 13 answered · rates start at 20
12.8 s
NVIDIA: Nemotron 3.5 Content Safety (free)nvidia/nemotron-3.5-content-safety:freevision · reasoningFree128K—No reports—
NVIDIA: Nemotron 3.5 Lightning (free)nvidia/nemotron-3.5-lightning:freetools · reasoningFree1M26.8
Too few7 of 14 answered · rates start at 20
52.4 s
Qwen: Qwen3.8 27B (free)qwen/qwen3.8-27b:freetools · vision · reasoningFree262K68.1
Too few0 of 8 answered · rates start at 20
9.1 s
Space Bunny Alphastealth/space-bunny-alphatools · vision · reasoningFree1M—No reports—
Thinking Machines: Inkling (free)thinkingmachines/inkling:freetools · vision · reasoningFree1.05M52.1
Too few0 of 7 answered · rates start at 20
7.6 s
Thinking Machines: Inkling Small (free)thinkingmachines/inkling-small:freetools · vision · reasoningFree1.05M52.9
Too few0 of 10 answered · rates start at 20
8.7 s

Prices are OpenRouter’s published text-token rates per million tokens; $0 isn’t unlimited. Reliability comes from signed cheapoS reports (1 member shares model details): a rate appears after 20 known outcomes, with its likely range. How we measure

How we measure

Published

Prices, context and capabilities come from the OpenRouter catalog, refreshed hourly (last Sep 29, 2026, 2:11 PM UTC). Prices are USD per million text tokens. A listing isn't a live availability check, and $0 comes with account rules and limits.

Benchmark

Coding, agentic and intelligence indices from Artificial Analysis, as distributed by OpenRouter. They measure test suites, not reliability on real work. Missing scores stay missing.

Measured

Accepted, signed request reports from cheapoS members who share model details. A rate appears only after 20 known outcomes, always with its likely range (95% Wilson interval); averages after 5 timed reports. “Answered” means the provider responded, not that the task succeeded. The sample is early and can be one person.

Before you pick a model

Are free models really free?

They list zero input and output token prices. You still need a provider account and API key, and usage limits apply. Optional tools or features may cost extra. See what free means at each provider.

Which free model is best for coding?

Filter for tool calling and sort by the coding benchmark as a starting point. Then check each model page for Club reports and their sample size, and try a small task in your own project. Benchmarks, request reliability and finished work answer different questions.

Why does a model have no Club reports?

It may be new, unused by members who share model details, or recorded under a route we can't match exactly. It stays listed because published availability is useful on its own. We never fill the gap with a guessed rate.

Can I help make the directory more useful?

Use cheapoS on real projects and opt in to Club sharing, including model details if you choose. Only accepted, signed reports count. Join the Club or read what gets shared.

Community compute notebookExact totals, role breakdown and shareable badge
Community compute · cheapos.lolReported community usage
250,750,936
Reported free & included tokens

Free intelligence, put to work.

Total: 1 public reporting members (up to 100). Breakdowns: all 2 sharing members. Models include shared names only.

Report fetched 2026-09-29 14:11:49 UTC. Updates when new reports arrive.

Estimated API equivalent
~$752.25

Illustrative reference: $3 USD per 1M reported tokens.

Free intelligence, real scale. This compares reported public-free, included-access and local tokens with a flat reference rate. Paid and unknown-access tokens are excluded.

An illustration, not actual charges or verified savings. Subscription, hardware and electricity costs are not deducted.

Prompts and code stay private. Installation signatures authenticate submitted reports; they do not independently audit provider billing.

Input vs output

Input is context sent to models, including repeated context. Output is what models generate. Token volume measures usage, not work quality.

Input
246,815,36698.4%
Output
3,935,5701.6%

7,422 reported requests across 1 reporting public profile. Public-free, included-access and local usage only.

Cached input & reasoning output
Cached input
27,254,813 tokens · 1,355 of 7,422 requests reported
Reasoning output
580,329 tokens · 1,321 of 7,422 requests reported

These reported subsets are already included in input or output. Missing details stay unknown; they are not added to the total.

Where it comes from250,750,936 categorized tokens in loaded profiles
Public free 248,713,832 (99.2%)
Included access 1,601,645 (0.6%)
Local models 435,459 (0.2%)
Roles in the mix · Reported role tokens
Worker
178,476,420
Input
176,783,951
Output
1,692,469
73.9% of reported role tokens
Reviewer
50,859,042
Input
49,216,215
Output
1,642,827
21.0% of reported role tokens
Planner
12,225,470
Input
11,860,762
Output
364,708
5.1% of reported role tokens
Coordinator
111,183
Input
107,308
Output
3,875
0.0% of reported role tokens
Shared model names:gemini-3.1-flash-lite 49,886,087openrouter/dots-studio/dots-3-note-preview:free 39,562,128groq/qwen/qwen3.6-27b 27,738,984antigravity/gemini-3.1-flash-lite 22,121,276openai/gpt-oss-120b 21,067,988+58 othersSVG badge ↗
RUN LOCALLY · EARLY ALPHA

Get cheapoS Free

Autonomous coding with local test execution and zero subscription tax. Choose your preferred environment:

Run via Git & Python (macOS / Linux)Requires Python 3.9+ and Git. No extra Python or JavaScript packages are needed to start the app:
git clone https://github.com/cheapos/CheapoS.git && cd CheapoS && python3 run.py

Opens the app in your browser at http://127.0.0.1:5173/. On macOS, you can also double-click Start CheapOS.command after cloning.

STEP 01

Grab 2 Free API Keys

cheapos harnesses generous free quotas. Get a free key from Google AI Studio (Gemini 2.5 Flash) and GroqCloud (Llama 3.3 70B). Zero credit card required.

STEP 02

Enter Any Task Prompt

Type what you want to build (CLI tool, micro-app, test suite). The fast worker drafts code while the reviewer critiques and fixes mistakes autonomously.

⚡ 0 bill shock · 100% local execution
STEP 03

Deterministic Test & Ship

cheapos never stops until local unit tests pass (e.g. 28/28 tests passing). Review the diff and 1-click share your finished product to the Workbench.

Explore 10 showcase builds →