Cheapest AI Model 2026
This guide ranks production AI models by cost and practical usefulness for real workloads.
It helps teams choose cheaper models for high-volume extraction, classification, content, routing and automation tasks.
The page supports Stob.AI buyers who want to reduce AI operating costs without damaging quality or reliability.
It connects model pricing decisions to the wider automation architecture needed for business systems.
Pricing updated 11 November 2026.
| Model | Provider | Input / 1M tokens | Cached input / 1M tokens | Output / 1M tokens | Context | Best for | Status | Notes |
|---|---|---|---|---|---|---|---|---|
| Claude Fable 5.2 | Anthropic | $8 | $0.2 | $40 | 1000000 | Complex coding, knowledge work and long-running agents | Generally available | Released 9 Nov. Standard $8 input / $40 output / $0.20 cache read per 1M tokens. Cache writes $10 for 5 minutes or $16 for 1 hour. Fast mode $16 input / $80 output; batch half standard. Anthropic reports up to 30% faster output than Fable 5.1. Text and image input, text output. Verified 11 Nov 2026. |
| Gemini 4 Carbon | $2 | $0.1 | $10 | 1048576 | Demanding multimodal analysis, coding and agent workflows | Generally available (paid tier) | Paid API GA 10 Nov. Standard $2 input / $10 output / $0.10 implicit cache per 1M through 30 Nov, then $4 / $20 / $0.20 from 1 Dec. No free API tier. Text, image, audio, video and PDF input, text output. Long decode uses the shared 1,048,576-token budget. Tool charges are separate. Verified 11 Nov 2026. | |
| Muse Ember | Meta | $0.1 | $0.004 | $0.5 | 1048576 | High-volume extraction, classification and bounded tool calls | Generally available (regional restrictions) | Released 9 Nov. Standard $0.10 input / $0.50 output / $0.004 cache read per 1M up to 200K input tokens. Above 200K, $0.30 / $1.50 / $0.012 applies to the whole request. Text, image, audio, video and PDF input, text output. Available in Meta Model API and all Muse Code plans. Verified 11 Nov 2026. |
| ERNIE 5.5 Flash Preview | Baidu | $0.28 | $0.028 | $1.13 | 1024000 | Testing long-context coding, research and tool-based workflows | Public preview (OpenRouter) | Listed 10 Nov. Baidu Qianfan route on OpenRouter: $0.28 input / $1.13 output / $0.028 cache read per 1M. These are OpenRouter rates. Maximum output limit unverified. Verified 11 Nov 2026. |
| Nex-N2.6 | Nex AGI | $0.12 | $0.024 | $0.4 | 262144 | Evaluating lower-cost tool-heavy workflows and coding agents | Hosted API (OpenRouter) | Listed 10 Nov. Nex AGI route on OpenRouter: current $0.12 input / $0.40 output / $0.024 cache read per 1M. Listed regular rates are $0.15 / $0.50 / $0.03. No EU in-region routing. Maximum output limit unverified. Verified 11 Nov 2026. |
| Kimi K3.1 Pro | Moonshot AI | $2 | $0.2 | $10 | 1048576 | Long-horizon coding, financial research, 3D creation, and complex agent workflows | Generally available | Standard $2 input / $10 output / $0.20 cached input per 1M. Cache-write fees depend on TTL. Always-on thinking. Maximum output parameter supports up to 1,048,576 tokens within the shared context budget, with 131,072 default; this is not an additional output window. Open weights available under the Kimi K3.1 Pro license. Highspeed is a separate SKU. Verified 5 Nov 2026. |
| MiniMax M3.1 Flash | MiniMax | $0.15 | $0.015 | $0.6 | 1000000 | Low-cost multimodal agents, coding, and computer-use workflows | Generally available | Released 19 Oct 2026. Flat standard $0.15 input / $0.60 output / $0.015 cached input per 1M. Text, image and video input; text output. Always-on thinking and tool calling. Published open weights under the MiniMax Community license. Maximum output limit not verified. Verified 5 Nov 2026. |
| DeepSeek V4.1 Flash (1017) | DeepSeek | $0.3 | $0.006 | $1.2 | 1000000 | Low-cost long-context vision, coding and agents; peak rates shown, off-peak rates are half | Generally available | Current DeepSeek-V4.1-Flash-1017, updated 17 Oct 2026; API ID deepseek-flash. Peak standard $0.30 input / $1.20 output / $0.006 cached input per 1M; off-peak $0.15 / $0.60 / $0.003. Peak hours 01:00-04:00 and 06:00-10:00 UTC on weekdays excluding Chinese public holidays. Native image input; thinking, JSON and tools. MIT open weights. Legacy V4 Flash IDs route to this model; V4 Pro remains separately available. Verified 5 Nov 2026. |
| Mistral Large 4 (Le Chonk) | Mistral AI | $0.7 | $0.07 | $2.1 | 1000000 | Open-weight multimodal coding, enterprise work, and controlled deployment | Generally available (open weights) | Le Chonk is the same model as Mistral Large 4. Improved open weights released 29 Oct 2026. Standard $0.70 input / $2.10 output / $0.07 cached input per 1M. Generally available since 5 Nov 2026. Low, medium and high reasoning effort; high default. Maximum output limit not verified. Mistral Ecosystem Open-Weight license has additional Model-as-a-Service conditions. Verified 5 Nov 2026. |
| Claude Sonnet 5.5 | Anthropic | $2 | $0.1 | $10 | 1000000 | Everyday coding, polished documents, spreadsheets, slides and business agents | Generally available | Released 28 Sep 2026. Standard $2 input / $10 output / $0.10 cached input per 1M. Cache writes $2.50 for 5 minutes or $4 for 1 hour. 128K synchronous output; 300K batch output is beta. Adaptive thinking; high effort default. Verified 5 Nov 2026. |
| Claude Haiku 5.5 | Anthropic | $0.1 | $0.01 | $0.5 | 1000000 | High-volume classification, extraction, summaries and subagents; shown rates apply to prompts up to 100K tokens | Generally available | Released 7 Oct 2026. Prices shown apply to prompts up to 100K tokens. Above 100K: $0.50 input / $0.05 cached input / $2.50 output per 1M. Cache writes $0.125 for 5 minutes or $0.20 for 1 hour below the threshold; $0.625 / $1 above it. 128K synchronous output; 300K batch output is beta. Adaptive thinking; medium effort default. Verified 5 Nov 2026. |
| Grok 4.7 | xAI | $2 | $0.5 | $6 | 500000 | Coding, agents and knowledge work; shown base rates apply up to 200K context | Generally available | Released 21 Sep 2026. Base standard $2 input / $6 output / $0.50 cached input per 1M. Requests above 200K context use higher rates; the fast variant is priced separately. Text and image input; tools and structured outputs. Maximum output limit not stated in checked model documentation. Verified 5 Nov 2026. |
| GLM-5.3 | Z.ai | $1.4 | $0.26 | $4.4 | 1000000 | Complex software engineering, long-horizon coding and production agents | Generally available | Standard $1.40 input / $4.40 output / $0.26 cached input per 1M. Text-only input and output. Always-on reasoning with low, high and max effort; max default. Cache storage currently free; web search fees separate. Published weights use the GLM-5.3 license with additional conditions for very large Model-as-a-Service businesses. Verified 5 Nov 2026. |
| GLM-5.3-Flash | Z.ai | $0.15 | $0.03 | $0.5 | 1000000 | Low-cost visual programming, multimodal agents and document workflows | Generally available | Standard $0.15 input / $0.50 output / $0.03 cached input per 1M. Text, image, video and file input; text output. Always-on reasoning and tool calling. MIT open weights. FlashX is a separate higher-speed SKU. Cache storage currently free; web search fees separate. Verified 5 Nov 2026. |
| Muse Wonder | Meta | $4 | $0.2 | $20 | 1048576 | Complex software engineering, long-running agents, subagent orchestration and computer use | Generally available (regional restrictions) | Released 12 Oct 2026. API ID muse-wonder-1.0. Standard $4 input / $20 output / $0.20 cached input per 1M; no long-context premium. Text, image, audio, video and PDF input; text output. Grounding fees separate. Regional restrictions apply. Maximum output limit not verified. Muse Wonder Full is a separate Muse Code-only offering. Verified 5 Nov 2026. |
| GPT-6.1 Terra | OpenAI | $1 | $0.05 | $5 | 1050000 | Fast visual reasoning, computer use, natural writing and everyday production agents | Generally available | Released 26 Oct 2026. 922,000 maximum input; cutoff 1 Jun 2026. Reasoning none–max; medium default. Tool calling requires Responses API. Standard cache writes $1.25/30 min or $2.50/12 hours per 1M. Above 272K input tokens, the whole request uses $2 input / $0.10 cached / $7.50 output, with cache writes doubled. Batch and Flex half Standard; Fast twice Standard; regional processing +10% where available; tool fees separate. Sol/Astra remain stronger choices for the hardest coding. Verified 27 Oct 2026. |
| Gemini 4 Argon | $2 | $0.1 | $10 | 1048576 | Enterprise knowledge work, long-context multimodal analysis and complex agents | Stable (paid tier) | Launch prices through 30 Nov 2026 only; from 1 Dec: $4 input / $20 output / $0.20 implicit cache per 1M. Output includes thinking. Batch and Flex $1/$5 at launch ($2/$10 from December); Priority $3.60/$18 ($7.20/$36 from December). Explicit caching and Live API unsupported. maxOutput shows synchronous limit; long decode can use up to the unified 1,048,576-token input/output budget. Search/Maps charges separate. Verified 14 Oct 2026. | |
| GLM-4.7-Flash | Z.ai | $0 | $0 | $0 | 200000 | Prototyping and zero-token-cost workloads that can tolerate free-tier limits | Free API tier | Official pricing lists input, cached input, cache storage, and output as free. Treat free capacity and rate limits separately from paid production SLAs. |
| Qwen3.7 Flash | Alibaba Cloud | $0.029668 | $0.005934 | $0.118671 | 1000000 | Ultra-low-cost multimodal chat, extraction, classification, and high-volume agents | Generally available | Standard columns are for inputs up to 32K. From 32K to 256K: CNY 0.6 input, 0.12 cache, 2.4 output. Above 256K: CNY 1.2, 0.24, 4.8. |
| Ministral 3 3B | Mistral AI | $0.1 | — | $0.1 | 256000 | Cheapest compact open-weight deployment and simple production tasks | Generally available | Input and output tokens have the same standard rate. Batch API is 50% below synchronous pricing. |
| Ministral 3 8B | Mistral AI | $0.15 | — | $0.15 | 256000 | Low-cost private deployment and compact multimodal applications | Generally available | Input and output tokens have the same standard rate. Batch API is 50% below synchronous pricing. |
| GLM-4.7-FlashX | Z.ai | $0.07 | $0.01 | $0.4 | 200000 | Very low-cost coding, chat, translation, and structured production workloads | Generally available | Cached input storage is listed as limited-time free. Web search is charged separately per use. |
| Gemini 2.5 Flash-Lite | $0.1 | $0.01 | $0.4 | 1048576 | Very low-cost classification, extraction, chat, and batch processing | Generally available | Standard columns use text, image, and video input rates. Audio input is priced separately. | |
| Ministral 3 14B | Mistral AI | $0.2 | — | $0.2 | 256000 | Private deployment, compact multimodal assistants, and edge-friendly workloads | Generally available | Input and output tokens have the same standard rate. Batch API is 50% below synchronous pricing. |
| Mistral Small 4 | Mistral AI | $0.15 | — | $0.6 | 256000 | Low-cost open-weight reasoning, coding, multimodal chat, and agents | Generally available | Batch API is 50% below synchronous token pricing. Cache savings can be up to 90%, with workload-dependent effective rates. |
| GPT-5.6 Luna | OpenAI | $0.2 | $0.02 | $1.2 | 1048576 | Low-cost production chat, extraction, classification, agents, and high-volume workloads | Generally available | Cheapest GPT-5.6 tier. Prompts above 272K tokens use long-context rates. |
| Qwen3.7 Plus | Alibaba Cloud | $0.296677 | — | $1.186709 | 1000000 | Cost-controlled multimodal agents, reasoning, and long-context applications | Generally available | Global list rate through 256K input tokens. Above 256K: CNY 6 input and CNY 24 output. Promotions, cache, and regional rates vary. |
| MiniMax M3 | MiniMax | $0.3 | $0.06 | $1.2 | 1000000 | Low-cost long-context multimodal coding and agent applications | Generally available | Standard columns use the permanent discounted rate through 512K tokens. Requests above 512K are priced at twice the standard rate. Priority tier costs more. |
| MiniMax M2.7 | MiniMax | $0.3 | $0.06 | $1.2 | 204800 | Affordable coding agents and general-purpose production automation | Generally available | Cache write price is separate. A high-speed variant is available at higher token rates. |
| Gemini 3.1 Flash-Lite | $0.25 | $0.025 | $1.5 | 1048576 | Budget multimodal applications and bulk processing | Generally available | Standard columns use text, image, and video input rates. Audio input and cache rates are higher. | |
| Mistral Large 3 | Mistral AI | $0.5 | — | $1.5 | 256000 | Open-weight general-purpose multimodal applications and self-hosting | Generally available | Batch API is 50% below synchronous token pricing. Cache savings can be up to 90%, with workload-dependent effective rates. |
| Gemini 3.5 Flash-Lite | $0.3 | $0.03 | $2.5 | 1048576 | Low-cost multimodal extraction, chat, and automation | Generally available | ||
| Gemini 2.5 Flash | $0.3 | $0.03 | $2.5 | 1048576 | Cost-effective multimodal assistants and high-throughput production | Generally available | Standard columns use text, image, and video input rates. Audio input is priced separately. | |
| Amazon Nova 2 Lite | Amazon | $0.3 | — | $2.5 | 1000000 | AWS-native multimodal assistants, document work, and cost-controlled reasoning | Generally available | Global model pricing shown. Batch and flex processing use lower rates. Regional and feature charges may vary. |
| GLM-4.7 | Z.ai | $0.6 | $0.11 | $2.2 | 200000 | Affordable coding, reasoning, research, and agent workflows | Generally available | Cached input storage is listed as limited-time free. Web search is charged separately per use. |
| Gemini 3.8 Flash | $0.75 | $0.075 | $3.75 | 1048576 | Fast multimodal agents, computer use, search-grounded apps and high-volume production | Generally available | Standard prices through 31 Dec 2026; from 1 Jan 2027: $1.50 input / $7.50 output / $0.15 cache. This is a calendar-based increase, not a 200K prompt surcharge. Batch and Flex are half standard price. Cache storage $0.50 per 1M tokens per hour through December, $1 from January. Verified 14 Oct 2026. | |
| Gemini 3.7 Flash | $0.75 | $0.075 | $3.75 | 1048576 | Low-latency multimodal agents, search-grounded apps, and high-volume workloads | Generally available | Listed promotional standard rate through 2026-12-31; scheduled to increase in 2027. | |
| Gemini 3.6 Flash | $0.75 | $0.075 | $3.75 | 1048576 | Multimodal production assistants and search-grounded workflows | Generally available | Listed promotional standard rate through 2026-12-31; scheduled to increase in 2027. | |
| Grok 4.3 | xAI | $1.25 | $0.2 | $2.5 | 1000000 | Lower-cost xAI workloads that benefit from a 1M-token context window | Active legacy | Active legacy model. Requests above 200K input tokens use higher long-context rates. |
| Kimi K2.7 Code | Moonshot AI | $0.95 | $0.19 | $4 | 262144 | Open-weight coding agents, repository work, and software automation | Generally available | Batch token costs are 60% of standard rates. |
| Kimi K2.6 | Moonshot AI | $0.95 | $0.16 | $4 | 262144 | Cost-effective multimodal agents, reasoning, and coding | Generally available | Batch token costs are 60% of standard rates. |
| GLM-5-Turbo | Z.ai | $1.2 | $0.24 | $4 | 200000 | Persistent tool-using agents and high-throughput long-chain workflows | Generally available | Cached input storage is listed as limited-time free. Web search is charged separately per use. |
| DeepSeek V4 Pro | DeepSeek | $1.32 | $0.044 | $3.96 | 1000000 | Higher-quality long-context reasoning, coding, and autonomous agents | Generally available | Peak standard rate shown. Off-peak rates apply outside 01:00-04:00 and 06:00-10:00 UTC. |
| Claude Haiku 4.5 | Anthropic | $1 | $0.1 | $5 | 200000 | Fast chat, extraction, classification, support, and lightweight agents | Generally available | Prompt cache writes cost 1.25x input for 5-minute cache and 2x for 1-hour cache. |
| Muse Spark 1.3 | Meta | $1.25 | $0.15 | $4.25 | 1000000 | Long-horizon agentic coding and multimodal automation on Meta-hosted infrastructure | Generally available | Current Muse Spark release. Standard commercial rate shown; a lower contributor SKU ($0.10 in / $0.20 out) carries different data-use terms and is not used for this production comparison. |
| Muse Spark 1.2 | Meta | $1.25 | $0.15 | $4.25 | 1000000 | Superseded by Muse Spark 1.3 - kept for version comparison | Public preview | Superseded by Muse Spark 1.3. Standard commercial rate shown. A lower contributor SKU has different data-use terms and is not used for this production comparison. |
| GLM-5.2 | Z.ai | $1.4 | $0.26 | $4.4 | 1000000 | Open-weight long-horizon coding, reasoning, and production agents | Generally available | Cached input storage is listed as limited-time free. Web search is charged separately per use. |
| Qwen3.8 Max | Alibaba Cloud | $1.780063 | $0.222508 | $5.34019 | 1000000 | Large-context multimodal reasoning, agents, and enterprise workflows | Generally available | Global list rate. Regional availability, web search, and batch pricing vary. Explicit cache read and batch file rates may differ by region. |
| Grok 4.6 | xAI | $2 | $0.5 | $6 | 500000 | Current xAI reasoning, multimodal agents, and real-time information workflows | Generally available | Requests above 200K input tokens use higher long-context rates. |
| Mistral Medium 3.5 | Mistral AI | $1.5 | — | $7.5 | 256000 | Open-weight frontier multimodal agents, coding, and enterprise deployment | Generally available | Batch API is 50% below synchronous token pricing. Cache savings can be up to 90%, with workload-dependent effective rates. |
| Gemini 3.5 Flash | $1.5 | $0.15 | $9 | 1048576 | High-quality multimodal reasoning and agent workflows | Generally available | ||
| Gemini 2.5 Pro | $1.25 | $0.125 | $10 | 1048576 | Stable multimodal reasoning, coding, and long-context analysis | Generally available | Prompts above 200K tokens use higher long-context rates. | |
| Claude Sonnet 5 | Anthropic | $2 | $0.2 | $10 | 1000000 | Balanced production agents, coding, support, and document workflows | Generally available | Prompt cache writes cost 1.25x input for 5-minute cache and 2x for 1-hour cache. |
| GPT-5.6 Terra | OpenAI | $2 | $0.2 | $12 | 1048576 | Production agents and reasoning with stronger cost control | Generally available | Prompts above 272K tokens use the long-context input and output rates. |
| Gemini 3.1 Pro Preview | $2 | $0.2 | $12 | 1048576 | Frontier multimodal reasoning evaluations and pre-production testing | Preview | Preview model. Prompts above 200K tokens use higher long-context rates. | |
| Kimi K3 | Moonshot AI | $3 | $0.3 | $15 | 1048576 | Frontier open-weight multimodal agents, coding, and deep reasoning | Generally available | K3 is a trillion-scale open-weight multimodal reasoning model. Tool charges, if used, are separate. |
| Claude Opus 5.5 | Anthropic | $4 | $0.2 | $20 | 1000000 | Long-running agentic coding, research, and professional knowledge work | Generally available | Released 22 Sep 2026. 20% cheaper per token than Opus 5 and about 40% cheaper to run because it uses fewer tokens per task. Cache reads $0.20 per 1M, 5-minute cache writes $5 per 1M, Batch API 50% off. Adaptive thinking is always on and defaults to medium effort. |
| Claude Opus 5 | Anthropic | $5 | $0.5 | $25 | 1000000 | Previous-generation Opus; superseded by Claude Opus 5.5 | Generally available | Prompt cache writes cost 1.25x input for 5-minute cache and 2x for 1-hour cache. |
| GPT-5.6 Sol | OpenAI | $4 | $0.4 | $20 | 1048576 | Highest-capability reasoning, agents, coding, and multimodal work | Generally available | Promotional pricing through 21 Nov 2026. Prompts above 272K tokens use the long-context input and output rates. |
| GPT-6.1 Sol | OpenAI | $2 | $0.1 | $10 | 1050000 | Near-Astra coding, computer use and professional work at one-fifth of Astra pricing | Generally available | Released 29 Sep 2026; replaces GPT-6 Sol at the same token price with cached input halved. Cache write $2.50 per 1M. Prompts above 272K input tokens bill at $4 input / $0.20 cached / $5 cache write / $15 output. Reasoning effort low to max. |
| GPT-6 Sol | OpenAI | $2 | $0.2 | $10 | 1050000 | Complex coding, demanding reasoning and agentic workflows | Superseded | Released 22 Sep 2026. Cache write $2.50 per 1M. Prompts above 272K input tokens bill at $4 input / $0.40 cached / $5 cache write / $15 output. Reasoning effort from none to max. |
| GPT-6 Luna | OpenAI | $0.1 | $0.01 | $0.5 | 1050000 | Classification, extraction, routing, lightweight agents and high-volume production | Generally available | Released 22 Sep 2026. Cheapest GPT-6 tier with the same 1.05M context and reasoning controls as Sol. Above 272K input tokens: $0.20 input, $0.02 cached, $0.75 output. |
| Gemini 3.8 Live | $0.75 | — | $4.5 | 131072 | Low-latency realtime voice agents and streaming multimodal dialogue | Generally available | Stable 15 Sep 2026. Text rates shown for comparability: audio input $3 per 1M ($0.005/min), audio output $12 per 1M ($0.018/min), image and video input $1 per 1M. No caching. An Extended Thinking variant shares the same rates. | |
| GPT-6 Astra | OpenAI | $10 | $1 | $50 | 1050000 | Most complex reasoning, research, coding, agents, and document creation | Generally available | Cache write is $12.50 per 1M tokens. Prompts above 272K input tokens are billed at 2x input and cache rates and 1.5x output rates for the entire request. Knowledge cutoff 30 April 2026. |
| Claude Fable 5.1 | Anthropic | $10 | $0.25 | $50 | 1000000 | Long-horizon agentic coding, research, and highest-quality knowledge work | Active (previous generation) | Same input and output price as Fable 5, but cache reads drop to $0.25 per 1M tokens (a quarter of the old rate), cutting typical costs by about 25% and agentic workloads by up to 45%. Cache writes are $12.50 (5m) and $20 (1h). Batch API is 50% off. |
| Claude Fable 5 | Anthropic | $10 | $1 | $50 | 1000000 | Previous-generation frontier Claude; superseded by Fable 5.1 | Generally available | Prompt cache writes cost 1.25x input for 5-minute cache and 2x for 1-hour cache. |