Cheapest LLM for AI agents in 2026
Prices last updated 2026-10-25. USD per 1M tokens, standard tier, short context. 32 models from 9 providers.
For an agent, the cheapest model is the one with the lowest cost per task, not the lowest input price. Agents make many calls, and output tokens usually cost 2-5x more than input tokens.
Price table (sorted by example cost per task)
Example task: 5 calls, each with 3,000 input and 800 output tokens, for a total of 15,000 in and 4,000 out. Cost = 15,000 x input / 1M + 4,000 x output / 1M. Caching and retries are not included.
| Model | Provider | Input $/1M | Output $/1M | Cached $/1M | Cost/task | Notes |
|---|---|---|---|---|---|---|
| GPT-6 Luna | OpenAI | $0.1 | $0.5 | $0.01 | $0.0035 | Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request (OpenAI model page). Derived long-context rate: $0.20 in / $0.02 cached / $0.75 out. |
| Claude Haiku 5.5 | Anthropic | $0.1 | $0.5 | $0.01 | $0.0035 | Prompts up to 100K tokens; $0.50/$2.50 above 100K ($0.05 cached). |
| Mistral Small 4 | Mistral | $0.15 | $0.6 | $0.015 | $0.0046 | Open-weights model (Apache 2.0). |
| MiniMax M3.1 Flash | MiniMax | $0.15 | $0.6 | $0.015 | $0.0046 | API model MiniMax-M3.1-Flash, standard tier, 1M-token context. Priority tier is 1.5x standard. |
| GPT-5.6 Luna | OpenAI | $0.2 | $1.2 | $0.02 | $0.0078 | |
| DeepSeek Flash (V4.1 Flash 1017) | DeepSeek | $0.3 | $1.2 | $0.006 | $0.0093 | API model deepseek-flash, served by DeepSeek-V4.1-Flash-1017. Peak-hour price; off-peak $0.15/$0.60. Open-weights model. Server-side Search/Fetch tool calls (e.g. via Claude Code or Codex) cost $0.75 per 1K requests starting Nov 1, 2026. |
| MiniMax M3 | MiniMax | $0.3 | $1.2 | $0.06 | $0.0093 | API model MiniMax-M3, standard tier, prompts up to 512K input tokens (price includes MiniMax's permanent 50% discount). Above 512K: $0.60/$2.40 ($0.12 cached). Priority tier is 1.5x standard. |
| Mistral Large 3 | Mistral | $0.5 | $1.5 | $0.05 | $0.0135 | API name mistral-large-2512 (docs page mistral-large-3-25-12). Previous Mistral Large generation. 256K context. Cached input $0.05 per docs.mistral.ai/inference/pricing (verified Oct 24, 2026). |
| Grok Build 0.1 | xAI | $1 | $2 | $0.2 | $0.0230 | Under 200K context; $2/$4 above. |
| Gemini 3.8 Flash | $0.75 | $3.75 | $0.075 | $0.0262 | Through Dec 31, 2026; $1.50/$7.50 from Jan 1, 2027 (cached input $0.075 to $0.15). | |
| Grok 4.3 | xAI | $1.25 | $2.5 | $0.2 | $0.0288 | Under 200K context; $2.50/$5 above. |
| Kimi K2.7 Code | Moonshot | $0.95 | $4 | $0.19 | $0.0302 | API model kimi-k2.7-code. 262,144-token context. The kimi-k2.7-code-highspeed variant costs $1.90/$8.00 (cached $0.38). Retirement: from October 31, 2026, K2 series models are no longer available to newly registered users (existing users not affected), and Moonshot plans to retire all K2 series models by the end of December 2026. Suggested migration: Kimi K3.1 Pro ($2/$10). |
| Kimi K2.6 | Moonshot | $0.95 | $4 | $0.16 | $0.0302 | API model kimi-k2.6. 262,144-token context. Retirement: from October 31, 2026, K2 series models are no longer available to newly registered users (existing users not affected), and Moonshot plans to retire all K2 series models by the end of December 2026. Suggested migration: Kimi K3.1 Pro ($2/$10). |
| DeepSeek V4 Pro | DeepSeek | $1.32 | $3.96 | $0.044 | $0.0356 | Peak-hour price; off-peak $0.66/$1.98. Open-weights model. Server-side Search/Fetch tool calls (e.g. via Claude Code or Codex) cost $0.75 per 1K requests starting Nov 1, 2026. |
| Meta Muse Spark 1.3 | Meta | $1.25 | $4.25 | $0.15 | $0.0357 | Standard mode; Contributor mode (muse-spark-1.3-contributor, your data may train Meta models) is $0.10/$0.20. No long-context premium. |
| Mistral Large 4 | Mistral | $1.36 | $4.18 | $0.14 | $0.0371 | API name mistral-large-4. Open-weights model (public preview). |
| Mistral Medium 3.5 | Mistral | $1.5 | $7.5 | $0.15 | $0.0525 | |
| Grok 4.7 | xAI | $2 | $6 | $0.5 | $0.0540 | Under 200K context; $4/$12 above. |
| GPT-6.1 Sol | OpenAI | $2 | $10 | $0.1 | $0.0700 | Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request (OpenAI model page). Derived long-context rate: $4 in / $0.20 cached / $15 out. |
| GPT-6 Sol | OpenAI | $2 | $10 | $0.2 | $0.0700 | Previous Sol release; GPT-6.1 Sol has the same input/output price with cheaper cached input ($0.10). |
| Claude Sonnet 5.5 | Anthropic | $2 | $10 | $0.1 | $0.0700 | |
| Gemini 4 Argon | $2 | $10 | $0.1 | $0.0700 | Promo price through Nov 30, 2026; $4.00/$20.00 from Dec 1, 2026 (cached input $0.10 to $0.20). | |
| Kimi K3.1 Pro | Moonshot | $2 | $10 | $0.2 | $0.0700 | API model kimi-k3.1-pro. 1,048,576-token context at one flat price. Cache writes billed separately: $2.00 (5-min TTL, default) or $4.00 (1-hour TTL) per 1M tokens. The kimi-k3.1-pro-highspeed variant costs $4/$20. |
| GPT-5.6 Terra | OpenAI | $2 | $12 | $0.2 | $0.0780 | |
| Kimi K3 | Moonshot | $3 | $15 | $0.3 | $0.1050 | API model kimi-k3. 1,048,576-token context at one flat price. Cache writes billed separately: $3.00 (5-min TTL, default) or $6.00 (1-hour TTL) per 1M tokens. |
| GPT-5.6 Sol | OpenAI | $4 | $20 | $0.4 | $0.1400 | Promotional pricing, available at least through November 21, 2026 (per OpenAI). Cache read $0.40, 30-minute cache write $5.00 (default), 12-hour cache write $10.00 per 1M tokens. Long context (>272K input tokens): $8 input / $0.80 cached / $10 cache write / $30 output. Batch $2/$10, Fast mode $8/$40. |
| Claude Opus 5.5 | Anthropic | $4 | $20 | $0.2 | $0.1400 | |
| Meta Muse Wonder 1.0 | Meta | $4 | $20 | $0.2 | $0.1400 | API model muse-wonder-1.0. Standard mode only (no Contributor mode). No long-context premium. |
| Kimi K3.1 Pro Highspeed | Moonshot | $4 | $20 | $0.2 | $0.1400 | API model kimi-k3.1-pro-highspeed. Faster variant of K3.1 Pro. 1,048,576-token context at one flat price. Cache writes billed separately: $4.00 (5-min TTL, default) or $8.00 (1-hour TTL) per 1M tokens. Cached input stays $0.20, same as standard K3.1 Pro. |
| GPT-6 Astra | OpenAI | $10 | $50 | $1 | $0.3500 | Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request (OpenAI model page). Derived long-context rate: $20 in / $2 cached / $75 out. |
| Claude Fable 5.1 | Anthropic | $10 | $50 | $0.25 | $0.3500 | |
| Claude Mythos 5.1 | Anthropic | $10 | $50 | $0.25 | $0.3500 | Limited availability: only for organizations verified through Anthropic's verification programs. |
To see what a fixed budget buys on each of these models, check how many tokens $20, $50 and $100 of API credit buy.
Key takeaways
- Cheapest per task: GPT-6 Luna and Claude Haiku 5.5 at $0.1 / $0.5, or $0.0035 per example task. Haiku's price applies up to 100K context only.
- Off-peak discount: DeepSeek Flash drops to $0.15 / $0.60 off-peak, and DeepSeek V4 Pro to $0.66 / $1.98.
- Mid-tier cluster at $2 / $10: GPT-6 Sol, GPT-6.1 Sol, Claude Sonnet 5.5, Gemini 4 Argon and Kimi K3.1 Pro all cost $0.07 per example task. Argon goes to $4 / $20 from Dec 1, 2026.
- Price changes on the calendar: Gemini 3.8 Flash goes to $1.50 / $7.50 from Jan 1, 2027.
- Caching matters most for agents: cached input is as low as $0.01 per 1M tokens (GPT-6 Luna, Claude Haiku 5.5).
- Benchmarked cost per task: in the Artificial Analysis Coding Agent Index 2.0, real coding-agent tasks cost $0.16 to $13.30. See our benchmark page.
Calculate your own agent
Your agent's steps, context growth and retries will change the ranking.
Sources (official pricing pages)
- OpenAI: https://developers.openai.com/api/docs/pricing
- Anthropic: https://platform.claude.com/docs/en/about-claude/pricing
- Google: https://ai.google.dev/gemini-api/docs/pricing
- xAI: https://docs.x.ai/developers/pricing
- DeepSeek: https://api-docs.deepseek.com/quick_start/pricing
- Mistral: https://docs.mistral.ai/inference/pricing
- Meta: https://dev.meta.ai/docs/pricing-rate-limits
- Moonshot: https://platform.kimi.ai/docs/pricing/chat
- MiniMax: https://platform.minimax.io/docs/pricing/overview
- Gemini 4 Argon launch and price change: Google blog