Best AI model for agents: how to choose
There is no single best AI model for agents; it depends on budget, input size and whether you need to self-host. Highest of these on the official Terminal-Bench 4.0 board: Claude Opus 5.5. Lowest listed price: Mercury 3. Largest context window: GPT-6.1 Terra. Open weights: MiniMax M3.1 Flash. Every pick follows one stated fact, linked to its source. We have not run these models.
Reported Updated · Markdown · All models
Pick by what limits you
Highest of these on the official Terminal-Bench 4.0 board
64.8% on the official Terminal-Bench 4.0 board (snapshot Oct 26, 2026), the highest of the agent models in our data that appear on it. Independent board, not our test. Source
Lowest listed price
$0.1 in / $0.35 out per 1M tokens (temporary price), the lowest of the generally available agent models listed here. Source
Largest context window
1,050,000 tokens, the largest published among the generally available agent models here (tied; newest shown). Source
Open weights
Newest generally available agent model here with downloadable weights (MiniMax Community License); $0.15 in / $0.6 out per 1M tokens on the vendor API. Source
Each pick follows one stated rule over our sourced data, so it moves when the data does. This is a way to choose, not a quality ranking: we have not run these models.
All models in scope (29)
Language models in our data released in the 120 days before Oct 29, 2026 that their vendors position for long-running, tool-using work. Our data is not every model on the market.
| Model | Price (listed) | Context | Licence | Input | Released | Status |
|---|---|---|---|---|---|---|
| Gemini 4 CarbonGoogle | $2 in / $10 out per 1M tokens (introductory price, end date not stated; list $4 / $20)source | Not published | Proprietarysource | Not published | Oct 28, 2026source | Limited access |
| GPT-6.1 TerraOpenAI | $1 in / $5 out per 1M tokenssource | 1.05M tokenssource | Proprietary (API)source | text, imagesource | Oct 26, 2026source | Generally available |
| Kimi K3.1 ProMoonshot AI | $2 in / $10 out per 1M tokenssource | 1.05M tokenssource | Not yet publishedsource | Not published | Oct 23, 2026source | Generally available |
| Kimi K3.1 Pro HighspeedMoonshot AI | $4 in / $20 out per 1M tokenssource | 1.05M tokenssource | Not published | Not published | Oct 23, 2026source | Generally available |
| Mercury 3Inception | $0.1 in / $0.35 out per 1M tokens (launch promo, end date not stated; list $0.2 / $0.7)source | 1M tokenssource | Proprietary (API)source | Not published | Oct 19, 2026source | Generally available |
| MiniMax M3.1 FlashMiniMax | $0.15 in / $0.6 out per 1M tokenssource | 1M tokenssource | MiniMax Community Licensesource | text, image, videosource | Oct 19, 2026source | Generally available |
| Gemini 4 ArgonGoogle | $2 in / $10 out per 1M tokens (introductory price until Nov 30, 2026; list $4 / $20)source | 1.05M tokenssource | Proprietary (API)source | text, image, video, audio, pdfsource | Oct 13, 2026source | Generally available |
| Muse WonderMeta | $4 in / $20 out per 1M tokenssource | 1.05M tokenssource | Proprietary (API)source | text, image, video, audio, pdfsource | Oct 12, 2026source | Public preview |
| Muse Wonder FullMeta | Not published | Not published | Proprietary (subscription)source | Not published | Oct 12, 2026source | Limited access |
| Claude Haiku 5.5Anthropic | $0.1 in / $0.5 out per 1M tokenssource | 1M tokenssource | Proprietary (API)source | text, imagesource | Oct 7, 2026source | Generally available |
| GPT-6.1 SolOpenAI | $2 in / $10 out per 1M tokenssource | 1.05M tokenssource | Proprietary (API)source | text, imagesource | Sep 29, 2026source | Generally available |
| Claude Sonnet 5.5Anthropic | $2 in / $10 out per 1M tokenssource | 1M tokenssource | Proprietary (API)source | text, imagesource | Sep 28, 2026source | Generally available |
| Claude Opus 5.5Anthropic | $4 in / $20 out per 1M tokenssource | 1M tokenssource | Proprietary (API)source | text, imagesource | Sep 22, 2026source | Generally available |
| GPT-6 SolOpenAI | $2 in / $10 out per 1M tokenssource | Not published | Proprietary (API)source | text, imagesource | Sep 22, 2026source | Generally available |
| Grok 4.7xAI | $2 in / $6 out per 1M tokenssource | 500K tokenssource | Proprietary (API)source | text, imagesource | Sep 21, 2026source | Generally available |
| Claude Fable 5.1Anthropic | $10 in / $50 out per 1M tokenssource | 1M tokenssource | Proprietary (API)source | text, imagesource | Sep 15, 2026source | Generally available |
| GPT-6 AstraOpenAI | $10 in / $50 out per 1M tokenssource | 1.05M tokenssource | Proprietary (API)source | text, imagesource | Sep 3, 2026source | Generally available |
| Gemini 3.8 FlashGoogle | $0.75 in / $3.75 out per 1M tokens (temporary price until Dec 31, 2026; list $1.5 / $7.5)source | 1.05M tokenssource | Proprietary (API)source | text, image, video, audio, pdfsource | Sep 2, 2026source | Generally available |
| Claude Mythos 5.1Anthropic | $10 in / $50 out per 1M tokenssource | 1M tokenssource | Proprietary (API)source | Not published | Sep 1, 2026source | Limited access |
| GLM-5.3-FlashZ.ai | Not published | Not published | Not published | Not published | Aug 26, 2026source | Generally available |
| GLM-5.3Z.ai | $1.4 in / $4.4 out per 1M tokenssource | 1.05M tokenssource | GLM-5.3 License (MIT-style plus a MaaS condition)source | Not published | Aug 14, 2026source | Generally available |
| Grok 4.6xAI | $2 in / $6 out per 1M tokenssource | 500K tokenssource | Proprietary (API)source | text, imagesource | Aug 12, 2026source | Generally available |
| Gemini Robotics ER 2Google | Not published | Not published | Proprietary (API)source | Not published | Jul 30, 2026source | Public preview |
| Claude Opus 5Anthropic | $5 in / $25 out per 1M tokenssource | 1M tokenssource | Proprietary (API)source | Not published | Jul 24, 2026source | Generally available |
| Gemini 3.5 Flash-LiteGoogle | Not published | 1.05M tokenssource | Proprietary (API)source | text, image, video, audio, pdfsource | Jul 21, 2026source | Generally available |
| Gemini 3.6 FlashGoogle | Not published | 1.05M tokenssource | Proprietary (API)source | text, image, video, audio, pdfsource | Jul 21, 2026source | Generally available |
| Kimi K3Moonshot AI | $3 in / $15 out per 1M tokenssource | 1M tokenssource | Kimi K3 License (custom)source | Not published | Jul 16, 2026source | Generally available |
| GPT-5.6 SolOpenAI | $4 in / $20 out per 1M tokens (promotional price, OpenAI says it is available at least through Nov 21, 2026; list price after the promo not stated)source | 1.05M tokenssource | Proprietary (API)source | text, imagesource | Jul 9, 2026source | Generally available |
| Grok 4.5xAI | $2 in / $6 out per 1M tokenssource | 500K tokenssource | Proprietary (API)source | text, imagesource | Jul 8, 2026source | Generally available |
FAQ
What is the best AI model for agents right now?
It depends on what limits you. Highest of these on the official Terminal-Bench 4.0 board: Claude Opus 5.5. Lowest listed price: Mercury 3. Largest context window: GPT-6.1 Terra. Open weights: MiniMax M3.1 Flash. Each pick names the fact behind it and links to the source.
Which agent model is cheapest?
Mercury 3: $0.1 in / $0.35 out per 1M tokens (temporary price). Next: Claude Haiku 5.5 ($0.1 in / $0.5 out per 1M tokens); MiniMax M3.1 Flash ($0.15 in / $0.6 out per 1M tokens).
Is there an open-weight model for agents?
Yes: MiniMax M3.1 Flash (MiniMax Community License), GLM-5.3 (GLM-5.3 License (MIT-style plus a MaaS condition)), Kimi K3 (Kimi K3 License (custom)).
Did you test these models as agents?
No. This page is Reported. Prices, context windows and licences come from the vendor pages linked in the table; the Terminal-Bench 4.0 score comes from the official board, not from us.
Related
Latest news briefs on these models
- OpenAI GPT-6.1 Terra: a $1/$5 computer-use model, and what the launch post leaves out (Reported; first seen 2026-10-26 19:05 UTC)
The Weekly Verdict · free
5 new-model use cases. Every Monday.
A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.
Free. Double opt-in. One click to unsubscribe.