Updated September 24, 2026

LLM API pricing — the ledger

Every major model's list price in one table: input, output, and the cached-input rate that decides what an agent actually costs. Below it: the same open weights priced across hosts, and the GPU rates underneath everything.

Cheapest serious models right now: gpt-6-luna $0.1/0.5 · GLM-5.3-Flash $0.15/0.5 · Mistral Small 4 $0.15/0.6. Want monthly totals instead? Run your workload in the calculator →

OpenAIgpt-6-lunaCache write $0.125.$0.1$0.5$0.010128k
Z.aiGLM-5.3-Flash$0.15$0.5$0.030128k
MistralMistral Small 4$0.15$0.6$0.015128k
MistralMinistral 3 (8B / 14B)3B: $0.10/$0.10; 14B: $0.20/$0.20.$0.15$0.15$0.015128k
Googlegemini-3.1-flash-liteAudio input $0.50.$0.25$1.5$0.0251M
Googlegemini-3.5-flash-lite$0.3$2.5$0.0301M
DeepSeekdeepseek-flash (V4.1-Flash) — peakOff-peak: $0.15 in / $0.60 out.$0.3$1.2$0.0061M
MistralCodestral$0.3$0.9$0.030256k
Z.aiGLM-5.3-FlashX$0.37$1.25$0.075128k
MistralMistral Large 3Startlingly cheap for the class.$0.5$1.5$0.050128k
Z.aiGLM-4.7$0.6$2.2$0.11128k
Googlegemini-3.8-flash→ $1.50 / $7.50 on 2027-01-01.$0.75$3.75$0.0751M
AnthropicHaiku 4.5$1$5$0.1200k
xAIgrok-build-0.1$1$2$0.2256k
Z.aiGLM-5$1$3.2$0.2128k
Alibabaqwen3-max$1.2$6—32k
xAIgrok-4.3$1.25$2.5$0.21M
xAIgrok-4.20Reasoning / non-reasoning / multi-agent variants.$1.25$2.5$0.21M
Alibabaqwen3.6-max-preview$1.3$7.8—128k
DeepSeekdeepseek-v4-pro — peakOff-peak: $0.66 in / $1.98 out.$1.32$3.96$0.0441M
Z.aiGLM-5.3$1.4$4.4$0.26200k
Z.aiGLM-5.2 / GLM-5.1$1.4$4.4$0.26200k
MistralGLM-5.3 (resold)$1.4$4.4—128k
Googlegemini-3.5-flash$1.5$9$0.151M
MistralMistral Medium 3.5$1.5$7.5$0.15128k
OpenAIgpt-6-solThe default agent pick. Cache write $2.50.$2$10$0.2400k
AnthropicSonnet 5Best cache-read value. Cache write $2.50.$2$10$0.2200k
xAIgrok-4.7$2$6$0.5500k
xAIgrok-4.6$2$6$0.5500k
xAIgrok-4.5$2$6$0.3500k
Alibabaqwen3.8-maxBeijing: $1.65 / $4.95.$2$6—256k
Alibabaqwen3.7-max$2.5$7.5—256k
Alibabaqwen3.8-max-prime (Beijing)$3.3$9.9—256k
OpenAIgpt-5.6-sol (cyber)Security-tuned line.$4$20$0.4128k
AnthropicOpus 5.5Cache write $5.$4$20$0.2200k
AnthropicOpus 5 / 4.8 / 4.7 / 4.6 (legacy)Still billed; not recommended.$5$25$0.5200k
OpenAIgpt-6-astraFlagship. Cache write $12.50.$10$50$1400k
AnthropicFable 5.1Top tier. Cache write $12.50.$10$50$0.25200k
OpenAIgpt-5.6-cyber$12.5$75$1.25128k

39 of 39 models · click a column to sort

Provider notes

OpenAI
The GPT-6 line. Standard tier, short context — long context doubles input and output, Fast mode doubles everything, Batch and Flex halve it. Cyber models bill separately. Regional (data-residency) endpoints add +10% for models released on/after 2026-03-05. [developers.openai.com/api/docs/pricing]
Anthropic
Fable 5.1 sits above Opus. Batch −50%; cache reads are the cheapest per token on the market; US-only inference +10%; Opus 5.5 fast mode 2×. Legacy Opus 4.6–5 still bills at $5/$25. [claude.com/pricing]
Google
Gemini 3.8 Flash is the new default; note the scheduled price increase on 2027-01-01 (shown in the notes). Batch/Flex −50%, Priority 1.8×. Explicit context caching bills storage at $0.50/Mtok/hour (doubling on 2027-01-01). [ai.google.dev/gemini-api/docs/pricing]
xAI
Grok 4.x. Above a 200k-token threshold, long-context rates are 2× the listed input and output. Cached input is aggressively priced. [docs.x.ai/developers/models]
Z.ai
GLM-5.3 family. Cached-input storage is listed as limited-time free; GLM-4.7-Flash and GLM-4.5-Flash are free. The coding-plan subscription (flat monthly) is metered separately from API usage. [docs.z.ai/guides/overview/pricing]
DeepSeek
Peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays; everything else is half price. 1M context, 384k max output. Current ids: deepseek-flash and deepseek-v4-pro. [api-docs.deepseek.com/quick_start/pricing]
Alibaba (Qwen)
Model Studio international (Singapore) region; the Beijing region bills lower in CNY terms. Batch −50%; the context-cache discount applies to input only and does not stack with batch. [alibabacloud.com/help/en/model-studio/model-pricing]
Mistral
European hosting; cached input −90%; regional inference +10%; Enterprise APIs +75%. Notably now resells GLM-5.3 as a third-party model at Z.ai's list price. [mistral.ai/pricing/api]

Same weights, different hosts

Open-weight models are free to download, so the price is whatever host you pick — and the spread between hosts is often larger than the spread between models. Input price per million tokens:

GLM-5.3

spread $0.56–$1.40 /Mtok in · 2.5× difference

Z.ai$1.4 in / $4.4 out
Mistral$1.4 in / $4.4 out
Together$1.4 in / $4.4 out
DeepInfra$0.563 in / $2.5 out

GLM-5.3-Flash

spread $0.07–$0.15 /Mtok in · 2.0× difference

Z.ai$0.15 in / $0.5 out
Together$0.15 in / $0.5 out
Fireworks$0.15 in / $0.5 out
DeepInfra$0.075 in / $0.25 out

DeepSeek V4.1 Flash

spread $0.14–$0.30 /Mtok in · 2.1× difference

DeepSeek (peak)$0.3 in / $1.2 out
Fireworks$0.22 in / $0.66 out
Together$0.3 in / $1.2 out
DeepInfra$0.14 in / $0.42 out

DeepSeek V4 Pro 0813

spread $0.66–$1.32 /Mtok in · 2.0× difference

DeepSeek (peak)$1.32 in / $3.96 out
Together$1.32 in / $3.96 out
DeepInfra$0.66 in / $2.6 out

Qwen3.5-397B-A17B

spread $0.45–$0.60 /Mtok in · 1.3× difference

DeepInfra$0.45 in / $3 out
Together$0.6 in / $3.6 out

Qwen3.8-2.4T-A95B

single host listed

DeepInfra$2 in / $6 out
Together$2 in / $6 out

gpt-oss-120b

single host listed

Groq$0.15 in / $0.6 out

gpt-oss-20b

single host listed

Groq$0.075 in / $0.3 out

Self-hosting instead of APIs? The full GPU rental rate card lives on its own page — what an H100, B200 or A100 costs per hour, and what each is for.

GPU pricing →

Reading the numbers honestly

Cached input is the real agent price. Agents resend their whole prefix every turn, and every major provider prices that repeated prefix at roughly a tenth of fresh input. A model that looks 3× pricier on paper can be cheaper in practice if its cache terms are better and your hit rate is high.

Output tokens dominate bills. A 10:1 input-output price ratio still costs more on output for any agent that writes code or long replies — output is where token counts and dollars concentrate.

List price is not your price. Batch APIs halve costs when you can wait, off-peak windows cut more, and committed-use discounts are negotiated quietly. Use this ledger to compare; use the provider's console to budget.

How this stays current: an automated Pocket agent (the Pricing Watch agent — sources listed per provider below) re-checks every pricing page weekly and files a change report; the table updates from those findings, and the date at the top is the last verified pass. That is the same machinery that writes our trend research.

Pay list price, on your own key

Pocket agents run on your API keys at provider list price — no token markup, ever.

See how BYOK works

Live · refreshed every 6 hours

Every model: 361 models, 58 labs

The verified ledger above plus 441 models priced live from OpenRouter's public catalogue. Search, filter by capability and sort by price or context.

361 matches · prices in USD per million tokens · blended = weighted by your input:output ratio

ModelInputOutputCachedBlendedContextCan
Flint (free)Recursivefreefree—free262Ktools · reasoning
MiniMax M3.1 FlashMiniMax$0.150$0.600$0.015$0.2621Mvision · tools · reasoning
DeepSeek V4.1 Flash 1017DeepSeek$0.150$0.600$0.0030$0.2621.0Mvision · tools · reasoning
GlimpseElorian$0.600$3.00$0.060$1.20262Kvision · tools · reasoning
Xing4.0 29B A4B (free)Xingchen Agifreefree—free262Ktools · reasoning
Beam 1.0 501B A23BReflection$0.250$1.00$0.050$0.4381.0Mtools · reasoning
Mercury 3Inception$0.100$0.350$0.010$0.1631Mtools · reasoning
Moondream 3 PreviewMoondream$0.300$2.50—$0.85032Kvision · reasoning
Moondream 3.1Moondream$0.300$2.50—$0.85032Kvision · reasoning
Muse Spark OpenMeta$1.25$4.25$0.150$2.001.0Mvision · audio · tools · reasoning
Nemotron 3.5 Super VLNVIDIA$0.150$0.500$0.020$0.2371.0Mvision · tools · reasoning
Pareto 26.10Unbiased$0.800$3.20$0.030$1.401.0Mvision · tools
Pareto 26.9Unbiased$2.50$7.50$0.250$3.75262Kvision · tools
LFM2.5-24B-A2BLiquid AI$0.050$0.700$0.025$0.212128Ktools · reasoning
Gemini 4 ArgonGoogle$2.00$10.0$0.100$4.001.0Mvision · audio · tools · reasoning
SummarizeScaledown$0.040free—$0.0301Mvision
ExtractScaledown$0.040free—$0.0301Mvision
Muse WonderMeta$4.00$20.0$0.200$8.001.0Mvision · audio · tools · reasoning
Step 5 PreviewStepFun$1.00$2.70$0.100$1.431.0Mvision · tools · reasoning
Claude Haiku 5.5Anthropic$0.100$0.500$0.010$0.2001Mvision · tools · reasoning
Nano Banana 2.1Google$1.50$7.50—$3.0066Kvision · images · tools · reasoning
Mistral Large 4Mistral$1.36$4.18$0.140$2.061.0Mvision · tools · reasoning
Ling 3.1 FlashinclusionAI$0.075$0.225$0.015$0.112262Ktools · reasoning
GPT-6.1 Sol ProOpenAI$2.00$10.0$0.100$4.001.1Mvision · tools · reasoning
GPT-6.1 SolOpenAI$2.00$10.0$0.100$4.001.1Mvision · tools · reasoning
Claude Sonnet 5.5Anthropic$2.00$10.0$0.100$4.001Mvision · tools · reasoning
Perceptron Mk1.5Perceptron$0.150$1.50—$0.48737Kvision · audio · tools · reasoning
Ember-1Fireworks$3.00$15.0$0.300$6.001.0Mvision · tools · reasoning
GLM 5.3 PrimeZ.ai$2.80$8.80$0.560$4.301Mtools · reasoning
qwen3.8-max-prime (Beijing)Alibaba · verified list price$3.30$9.90—$4.95256Kvision · tools · reasoning
Aion 3.5 MiniAion Labs$0.700$1.40$0.180$0.875262Ktools · reasoning
Aion 3.5Aion Labs$3.00$6.00$0.750$3.75262Ktools · reasoning
Solar Mini 4Upstage$0.050$0.200$0.0050$0.088524Ktools · reasoning
Command A+Cohere$0.300$1.50$0.150$0.600192Kvision · tools · reasoning
GPT-6 Luna ProOpenAI$0.100$0.500$0.010$0.2001.1Mvision · tools · reasoning
gpt-6-lunaOpenAI · verified list price$0.100$0.500$0.010$0.200128Kvision · tools · reasoning
GPT-6 Sol ProOpenAI$2.00$10.0$0.200$4.001.1Mvision · tools · reasoning
gpt-6-solOpenAI · verified list price$2.00$10.0$0.200$4.00400Kvision · tools · reasoning
Claude Opus 5.5Anthropic$4.00$20.0$0.200$8.001Mvision · tools · reasoning
MiMo-V2.6-Pro-UltraSpeedXiaomi$4.35$8.70$0.036$5.441.0Mvision · audio · tools · reasoning
MiMo-V2.6-FlashXiaomi$0.140$0.280$0.0028$0.1751.1Mvision · audio · tools · reasoning
MiMo-V2.6-ProXiaomi$0.435$0.870$0.0036$0.5441.1Mvision · audio · tools · reasoning
grok-4.7xAI · verified list price$2.00$6.00$0.500$3.00500Kvision · tools · reasoning
Qwen3.8 Omni FlashAlibaba Qwen$0.150$0.470$0.016$0.2301Mvision · audio · tools · reasoning
Ternary Bonsai 2 27BPrism Ml$0.075$0.500$0.037$0.181262Kvision · tools · reasoning
GLM-5.3-FlashXZ.ai · verified list price$0.370$1.25$0.075$0.590128Kvision · tools · reasoning
Schematron V2 TurboInference Net$0.030$0.150$0.030$0.060128K
Schematron V2 SmallInference Net$0.050$0.230$0.050$0.095128K
Fugu Ultra v2Sakana$5.00$30.0$0.500$11.31Mvision · tools · reasoning
Fugu MaxSakana$2.00$6.00$0.250$3.001Mvision · tools · reasoning
Ling 3.0 Flash VLinclusionAI$0.060$0.180$0.012$0.090131Kvision · tools · reasoning
DeepSeek V4.1 FlashDeepSeek$0.165$0.660$0.0060$0.2891.0Mvision · tools · reasoning
Mercury 2.5Inception$0.040$0.150$0.0040$0.068260Ktools · reasoning
Nex-N2.5-MiniNex Agi$0.025$0.100$0.0025$0.044262Kvision · reasoning
Nex-N2.5-ProNex Agi$0.075$0.250$0.015$0.119262Kvision · tools · reasoning
gpt-6-astraOpenAI · verified list price$10.0$50.0$1.00$20.0400Kvision · tools · reasoning
GPT-6 Astra ProOpenAI$10.0$50.0$1.00$20.01.1Mvision · tools · reasoning
qwen3.8-maxAlibaba · verified list price$2.00$6.00—$3.00256Kvision · tools · reasoning
Muse Spark 1.3 ContributorMeta$0.100$0.200$0.0020$0.1251.0Mvision · tools · reasoning
Muse Spark 1.3Meta$1.25$4.25$0.150$2.001.0Mvision · tools · reasoning