by Nex AGI
Nex-N2.6 pricing, plans and limits
The newest agent model from Nex AGI, positioned by OpenRouter's listing as a step on Nex-N2.5-Pro for tool-following.
- Best for
- Computer-use and browser agents on a hosted API
- Not for
- Work that needs open weights
Should you use Nex-N2.6?
Nex AGI listed Nex-N2.6 on OpenRouter on 10 November 2026 as a model built on GLM-5.3-Flash with about 321 billion parameters, and OpenRouter measured 80 tokens per second at 0.89 seconds of latency. It suits teams building computer-use or coding agents on a hosted API, though it has no published benchmarks or public weights yet.
Best for
- Computer-use and browser agents on a hosted API Image input, tool calling and a reasoning-effort setting are all listed on OpenRouter, and the listing text names computer use as a target.
- Tool-heavy coding loops OpenRouter's listing says it follows the environment's tools more closely than Nex-N2.5-Pro.
Not for
- Work that needs open weights OpenRouter's weights link led to a missing page, and no repository is listed in the Nex AGI Hugging Face organisation.
- Jobs that need a million-token window The window is a quarter of the one on its GLM-5.3-Flash base.
- Decisions that need published benchmarks Nex AGI has published no scores for this model on any page read.
Can you use Nex-N2.6?
- OpenRouter (Nex AGI host): yes (Pay per token; one host, Nex AGI). Listed as Nex-N2.6.
- Free plan: not published.
Checked 11 Nov against Nex AGI's own pages.
Nex-N2.6 pricing
- Input, per 1M tokens
- $0.12/1M
- Output, per 1M tokens
- $0.40/1M
- Cached input, per 1M tokens
- $0.02/1M
- Pricing model
- per-token
- Free tier
- No
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.0036 | $0.0004 | $0.0040 |
| Support reply | $0.0002 | $0.0001 | $0.0004 |
| One coding agent run | $0.024 | $0.0080 | $0.032 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Nex AGI is the only host listed on OpenRouter. The 20 percent discount shown there has no published end date (read November 2026); the pricing table lists both the discounted and the list rate.
What Nex-N2.6 can do
Nex-N2.6 is an agentic text-and-image-input model from Nex AGI, built on GLM-5.3-Flash, with a 262,144-token context window and up to 235,929 tokens of output. Per OpenRouter's listing text it targets computer use, coding and long-horizon tool tasks, and it reduces repetitive reasoning compared with Nex-N2.5-Pro.
What goes in and what comes out
Reads
- text
- image
- tool-calls
Writes
- text
- tool-calls
Capabilities
- Function callingSupported
- Structured outputSupported
- Tool useSupported
- VisionSupported
- Audio inputNot supported
- Audio outputNot supported
- Video inputNot supported
- Reasoning modes
- configurable reasoning effort
Key features
Log probabilities and top_k
OpenRouter's parameter list includes logprobs, top_logprobs and top_k, which help with evaluation and sampling control.
Reasoning traces on request
include_reasoning is accepted, so reasoning output can be returned alongside the answer.
Separate cache-read rate
OpenRouter's provider table lists a distinct, lower rate for cache reads than for fresh input.
Earlier models on the same API
Nex AGI's earlier Mini and Pro models are listed under the same nex-agi namespace, so a team can compare generations through one API.
Known limitations
- No benchmark scores, weights, data policy or safety card found, so quality is unproven.
- Context is a quarter of the window Hokai records for its GLM-5.3-Flash base.
Who it is for
- Developers building computer-use or browser agents on a hosted API
- Teams already routing through OpenRouter that want an agent-tuned model
- Evaluators following Nex AGI's open agent models
Nex-N2.6 benchmarks
Standout result
Parameters, per OpenRouter's listing text
Source: OpenRouter model page and API, read 11 Nov 2026Where it sits
| Measure | This model | Rank | Median |
|---|---|---|---|
| Blended price (3:1)$ per 1M tokens | $0.19/M | #16 / 103 | peer median $1.71/M |
Ranks are against GA models on Hokai that publish the same figure; ties share a rank.
Strengths and limits
Strengths
- Tool-following is the stated gain: OpenRouter's listing says it sticks closer to an environment's tool list than Nex-N2.5-Pro, which matters for agent loops.
- Image input plus tool calling in one hosted model suits computer-use agents.
- First-party hosting: OpenRouter shows Nex AGI itself as the provider, so there is no reseller in between.
- OpenRouter's listing says it follows the tools available in the current environment more closely than Nex-N2.5-Pro.
- Accepts image input alongside text, which suits computer-use agents.
Limits
- No benchmark scores, system card or data policy were found, so quality is unproven.
- No public weights: the weights link on OpenRouter was a dead end when checked.
- OpenRouter shows a single host, so there is no provider failover on that route.
Nex-N2.6 alternatives
| Model | Lab | Context | In / out per 1M | Reads and writes |
|---|---|---|---|---|
| Nex-N2.6 This page | Nex AGI | 262K | $0.12 / $0.40 | text, image → text |
 GLM-5.3-FlashPick Nex-N2.6 for agent tuning and image input on a hosted API; pick GLM-5.3-Flash for open weights and a 1,048,576-token window against Nex-N2.6's 262,144. | Z.ai | 1M | $0.15 / $0.50 | text, image, video, pdf → text |
 MiMo-V2.6-FlashPick Nex-N2.6 if you want a first-party agent model with tool-following tuning; pick MiMo-V2.6-Flash for open weights and a window of about a million tokens. | Xiaomi Corporation | 1M | $0.14 / $0.28 | text, image, video, audio → text |
 Claude Fable 5Superseded by Fable 5.1 and Fable 5.2 but still sold on the Claude API: Anthropic's first generally available Mythos-class model, released June 2026, with a 1M-token context. | Anthropic | 1M | $10 / $50 | text, image, pdf → text |
Nex-N2.6 specs
- Context window
- 262K tokens
- Max output
- 236K tokens
- Latency (P50)
- 890 ms
- Tokens per second
- 80 tok/s
- Mixture of experts
- No
- Parameters
- About 321B (per OpenRouter's listing)
- Architecture
- Post-trained from GLM-5.3-Flash (per OpenRouter's listing)
API endpoints
| Name | Endpoint | Auth |
|---|---|---|
| OpenRouter | https://openrouter.ai/nex-agi/nex-n2.6 | API key |
Known quirks
Show 3 known quirks
Discount with no end date
OpenRouter shows a 20% discount next to a struck-through list rate and publishes no end date for it.
Workaround: Budget at the list rate and treat the discount as a bonus.
Weights link does not resolve
OpenRouter links a nex-agi repository on Hugging Face that returned a missing-page response on 11 November 2026.
Workaround: Use the hosted API, or the open Nex-N2.5 checkpoints if you need weights.
Smaller window than its base
The API lists 262,144 tokens of context while Hokai records 1,048,576 for GLM-5.3-Flash.
Workaround: Split long inputs, or call the base model for jobs that need the larger window.
Nex-N2.6 facts
- Deployment
- API · OpenRouter
- Released
- 10 Nov 2026
- Current version
- Nex-N2.6
- Family
- Nex-N2
- Variant
- Multimodal
- Licence
- Proprietary (hosted API; no public weights found)
Version history
Nex-N2.6 · 10 Nov 2026
Listed on OpenRouter and hosted by Nex AGI.
Nex-N2.5-Pro · 08 Sep 2026
Predecessor named in OpenRouter's listing text.
What changed
- Release Nex-N2.6 listed on OpenRouter source
Questions about Nex-N2.6
What does Nex-N2.6 cost per million tokens?
OpenRouter lists Nex AGI as the only host, read on 11 November 2026: $0.12 per 1M input tokens and $0.40 per 1M output tokens after a 20% discount, against a list price of $0.15 and $0.50. Cached input is $0.024 at the discount and $0.03 at list. The page gives no end date for the discount.
Can I download Nex-N2.6 or run it locally?
Not as far as Hokai could find. OpenRouter links a nex-agi repository on Hugging Face for the weights, which returned a missing page on 11 November 2026. Nex AGI's earlier Nex-N2.5 checkpoints are public with an Apache 2.0 tag, and no free tier was listed for Nex-N2.6.
How is Nex-N2.6 different from Nex-N2.5-Pro?
OpenRouter's listing text says Nex-N2.6 cuts repetitive reasoning and sticks closer to the environment's tool list than Nex-N2.5-Pro does. It also says Nex-N2.6 is built on GLM-5.3-Flash, whereas Nex AGI's N2.5 page describes Mini and Pro as building on the Nex-N2 multimodal base. No benchmark numbers exist for N2.6 to measure the difference.
What is Nex-N2.6 built on, and how long can its prompts be?
Per OpenRouter, the base is GLM-5.3-Flash and the size is roughly 321 billion parameters. Its API lists a 262,144-token context window and a 235,929-token maximum completion, well under the window of its GLM-5.3-Flash base.
Does Nex-N2.6 accept images and structured output?
Yes, per OpenRouter's API read on 11 November 2026: it takes text and image input, returns text, and accepts tools, tool_choice, a JSON-schema response format and a reasoning setting with an effort level. Hokai found no audio or video input.
About Nex-N2.6
Nex-N2.6 is an agentic model from Nex AGI, listed on OpenRouter on 10 November 2026. OpenRouter's listing text says it is built on GLM-5.3-Flash, a model from Z.ai, with about 321 billion parameters and is suited to computer use, coding and long-horizon tasks that involve tools. Compared with Nex-N2.5-Pro, the same text says, it reduces repetitive reasoning and follows the tools available in the current environment more closely. Hokai found no Nex AGI page for the model, so these descriptions are as relayed by OpenRouter.
Read the full background
No benchmark scores for Nex-N2.6 appeared on any page Hokai read, so the record carries none. Nex AGI's Nex-N2.5 page charts the earlier models against Claude Opus 5 and other rivals, but those numbers describe the N2.5 line, not this model. OpenRouter measured a median of 80 tokens per second and 0.89 seconds of latency on 11 November 2026, on the only host, which is Nex AGI itself.
The context window is 262,144 tokens and the maximum completion is 235,929 tokens, per OpenRouter's API. That is a quarter of the 1,048,576-token window Hokai records for GLM-5.3-Flash, so Nex-N2.6 does not pass the base model's larger window on to its users.
The model takes text and images in and gives text out. OpenRouter's API lists tools and tool_choice for function calling, JSON-schema structured outputs, a reasoning setting with an effort level and log probabilities. Hokai found no audio or video input.
Nex-N2.6 is served by Nex AGI through OpenRouter, and no other host is listed. OpenRouter's weights link, which points to a nex-agi repository on Hugging Face, led to a missing page on 11 November 2026, while Nex AGI's earlier Nex-N2.5 checkpoints carry an Apache 2.0 tag in the Hugging Face API. Treat Nex-N2.6 as API-only until a repository appears.
Choose GLM-5.3-Flash or MiMo-V2.6-Flash instead if you need open weights today or a window of about a million tokens, and look at Hy3 as a further open option. Teams that need published benchmark evidence should wait, because Nex AGI has released none for this model.
Nex AGI publishes no training data summary, cutoff, data policy or safety card for Nex-N2.6 on the pages Hokai read. The company record lists what the group has and has not disclosed.
Sources and guides
Official sources
- Nex-N2.6 model page Vendor
Specifications on this page (context window, token prices, benchmark scores, licence) are read from Nex AGI's own model card and documentation. The verdict is Hokai's reading of them, not the provider's claim.
Lists that include Nex-N2.6
The provider
More AI Models on Hokai
- Claude Fable 5.1Superseded by Claude Fable 5.2 in November 2026 but still sold on the Claude API: Anthropic's second Mythos-class model, released Sept 1, 2026, for long-horizon coding and research.
Gemini Decision 1Google DeepMind's decision model scores many questions over one multimodal input of up to 262,144 tokens and bills zero output tokens.
Shieldstral 1.0Mistral's open-weight multimodal safety classifier, released 4 August 2026, that screens text and images against policies you write.
- Claude Mythos 5.2Anthropic's November 2026 Mythos model: Fable 5.2's weights with relaxed cyber and biology safeguards for verified teams, with a 1M-token context.
ERNIE-5.5-Flash-PreviewBaidu's text-only preview mixture-of-experts model, served through Qianfan on OpenRouter with a 1M-token context window and tool calling.
Gemini 4 CarbonGoogle DeepMind's Gemini 4 Carbon, open to paying Gemini API developers since 10 November: a reasoning model for agentic coding, research and cyber defense.