Live survey286 models · 13 providers · 17,592 auditsAudited Using Git Worktrees · SAFE →What changedHow we verifyllms.txt

The Atlas · column B (<$0.10) · row 4 (256K–1M)

GRIDB4 Models at <$0.10 per 1M input with 256K–1M tokens of context

7 models · 3 creatorsChecked 10 Oct 2026

Open the map
7 models cost between $0.06 and $0.09 per 1M input tokens and hold 256K–1M tokens of context. The cheapest on input is Nemotron 3 Nano 30B A3B at
$0.06
observed We measured
Source
Our catalogue
Retrieved
10 Oct 2026
; Gemma 4 26B A4B has the lowest output price at $0.23. A request that fills each model’s window costs $0.02–$0.03 here, because windows differ within a square.
Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
CoCopilotAIGoogle AI Mode
Cite

Cite this page

ModelsAtlas. “Models at <$0.10 per 1M input with 256K–1M tokens of context”. Retrieved 10 Oct 2026. https://modelsatlas.com/atlas/b4

Models in B4

Cheapest available standard-tier offer · USD per 1M tokens

ModelInputOutputContextListed
Nemotron 3 Nano 30B A3BNVIDIA$0.06$0.24262,144—Alternatives
Gemma 4 26B A4BGoogle$0.07$0.23262,144free tier Alternatives
Qwen3 Coder 30B A3B InstructQwen (Alibaba)$0.07$0.28262,144—Alternatives
Nemotron 3 SuperNVIDIA$0.08$0.45262,144free tier Alternatives
Qwen3.5-35B-A3BQwen (Alibaba)$0.08$0.75262,144—Alternatives
Gemma 4 31BGoogle$0.09$0.34262,144free tier Alternatives
Qwen3 235B A22B Instruct 2507Qwen (Alibaba)$0.09$0.55262,144—Alternatives
Highlighted: lowest in column. “Listed” is what our records state: modalities, and whether a batch or free offer exists.

What a window-filling request costs here

One prompt that fills the model’s context window plus up to 4,000 output tokens · Σ derived

Nemotron 3 Nano 30B A3B262K window
$0.02
Gemma 4 26B A4B262K window
$0.02
Qwen3 Coder 30B A3B Instruct262K window
$0.02
Nemotron 3 Super262K window
$0.02
Qwen3.5-35B-A3B262K window
$0.02
Gemma 4 31B262K window
$0.02
Qwen3 235B A22B Instruct 2507262K window
$0.03

Arithmetic from the listed standard-tier rates, not a quote; the cheapest list price is not always the cheapest full-window request.

Neighbouring squares

One step cheaper, pricier, smaller or larger

← cheaper0GRIDA4No model in A4 yet.
pricier →13GRIDC4$0.10–0.30 per 1M input, 256K–1M context: Qwen3 30B A3B Instruct 2507, Qwen3 Next 80B A3B Instruct and 11 more.↓ smaller11GRIDB3<$0.10 per 1M input, 128–256K context: gpt-oss-20b, Mistral Nemo and 9 more.↑ larger4GRIDB5<$0.10 per 1M input, 1M context: Qwen3.7 Flash, Qwen3.5-Flash and 2 more.

Questions about square B4

What is the cheapest AI model with 256K–1M tokens of context at under $0.10 input?
Nemotron 3 Nano 30B A3B at $0.06 per 1M input tokens and $0.24 output. The other 6 in B4: Gemma 4 26B A4B, Qwen3 Coder 30B A3B Instruct, Nemotron 3 Super, Qwen3.5-35B-A3B, Gemma 4 31B and 1 more.
Which B4 model is cheapest for a request that fills its context window?
Nemotron 3 Nano 30B A3B, at about $0.02 for a prompt that fills its 262K window plus its output. The dearest of the 7 is Qwen3 235B A22B Instruct 2507 at $0.03. The cheapest list price is not always the cheapest full-window request, because windows differ within a square.
Are any B4 models open-weight?
None of the 7 is marked open-weight in our records.
What does grid reference B4 mean?
B is the price column (<$0.10 per 1M input tokens) and 4 is the context row (256K–1M). Models in the same square are like-for-like on price and context only; the grid says nothing about quality.

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.