GPT-6.1 Sol, Gemini 4 Argon, Nano Banana 2.1 and ElevenLabs voices: what is new on NetRoom

The short version

The NetRoom catalog gets several updates at once. In text: GPT-6.1 Sol and its Pro mode, Claude Sonnet 5.5, Claude Haiku 5.5, Google's new flagship Gemini 4 Argon, plus fresh Flash versions of DeepSeek and Qwen. Images gain Nano Banana 2.1 and Seedream 5.0 Flash, video gains three models, and the Sound section picks up a whole ElevenLabs family, two HeyGen voices and Fish Audio Drama 3. Every voice in the picker now also has a play button.

Some older models have left the catalog, and their pages lead to replacements. Below is what exactly changed and what to reach for per task. Prices are not repeated here: they live on the model cards, along with a calculator.

Text

GPT-6.1 Sol and Sol Pro

GPT-6.1 Sol replaces GPT-6 Sol. It is the upgrade one step below the GPT-6 Astra flagship, and it is built for working scenarios: agentic coding, computer use, document-heavy work and business workflow automation. The context window is 1.05M tokens, enough for a thick contract with its appendices or a sizable part of a codebase in a single request.

What changed against GPT-6 Sol: fewer factual errors, and explicit restrictions in the request are respected more reliably. If you write "do not touch these files" or "answer only from this document", the model holds the line better now.

GPT-6.1 Sol Pro is the same model in pro reasoning mode, for hard tasks where a mistake is expensive. The cost is time: Pro is slower and spends far more tokens on reasoning. Start with regular Sol and switch to Pro where the answer has to survive scrutiny.

Claude Sonnet 5.5 and Haiku 5.5

Claude Sonnet 5.5 takes over from Sonnet 5. It suits building features and fixing bugs, and also polished documents, slides and spreadsheets. It writes more clearly than its predecessor. Thinking is always on, so the effort setting becomes the main lever: it decides how much the model thinks about a task.

Claude Haiku 5.5 replaces Haiku 4.5. This is the small fast model for high-volume work: summaries, subagents, browser use. It is stronger at coding and computer use, has a 1M-token context, and is the first Haiku with adjustable effort. The price for users stayed the same as for Haiku 4.5.

Gemini 4 Argon

Gemini 4 Argon is Google's new flagship reasoning model. Its territory: complex engineering, math and science, and autonomous agents. It is also worth noting how it handles long documents that mix text, tables, charts and images: the model turns them into a single output grounded in the source material. On the Vals Index, Argon leads across finance, legal, coding and tax work.

It replaces Gemini 3.1 Pro: the old page now leads to the new one.

DeepSeek v4.1 Flash and Qwen3.8 Flash

DeepSeek v4.1 Flash and Qwen3.8 Flash take the places of DeepSeek v4 Flash and Qwen3.6 Flash. Both come from the "fast" lines, and both are on the list of free chat models below.

What is free in chat

Four text models now come with daily free requests: GPT-6 Luna, Gemini 3.5 Flash Lite, DeepSeek v4.1 Flash and Qwen3.8 Flash. A good way to compare them on your own tasks without topping up the balance.

Images

Nano Banana 2.1

Nano Banana 2.1 is Google's newest Nano Banana. It both generates and edits images at 1K, 2K and 4K, supports aspect ratios from 1:8 to 8:1 and accepts up to 14 reference images. It is better than earlier versions at placing a product into a new scene, and it renders text inside images with confidence.

The older Nano Banana, Nano Banana 2 and Nano Banana Pro now lead to the new model's page. Nano Banana 2 Lite stays in the catalog and keeps its free daily images.

Seedream 5.0 Flash

Seedream 5.0 Flash is the fast, cost-efficient tier of ByteDance's Seedream 5.0. It offers 1K and 2K output and up to ten reference images. It fits when you need to run through many variants without waiting on each one.

Video

Grok Imagine Video 1.5 Lite

Grok Imagine Video 1.5 Lite is a faster, lower-cost version of Grok Imagine Video 1.5. Modes: text-to-video and image-to-video from a first frame. Resolutions are 480p, 720p and 1080p, duration runs from 1 to 15 seconds, and audio is generated together with the picture.

MiniMax H3 Fast and H3 Max Turbo

MiniMax H3 Fast is the fast, low-cost H3. It outputs 480p, with durations from 4 to 15 seconds with sound, first and last frame control, and up to nine reference images or three reference videos.

MiniMax H3 Max Turbo is the faster H3 Max: 480p and 768p, clips from 5 to 15 seconds with sound, from text or from first and last frames.

If you need to lean on your own material, look at Fast: it takes reference images and videos. If you need a clip from a description or from two frames, Max Turbo fits.

Voice

The Sound section has grown noticeably. Before the models, the most practical change: every voice in the picker now has a play button with a short sample. You can hear a voice before generating instead of guessing from its name.

ElevenLabs: seven models

The ElevenLabs family has arrived: Eleven v4, Eleven v4 Turbo, Eleven v3, Multilingual v2, Flash v2.5, Turbo v2.5 and Turbo v2. There are 21 voices and a set of controls: stability, similarity, style, speaker boost, text normalization and language. Speed control is available on Flash, Turbo and Multilingual.

The models differ mostly in languages and expressiveness.

• Eleven v4 covers 90+ languages, Eleven v3 covers 70+
• v3 and v4 understand audio tags inside the text, such as [whispers] or [laughs], which set whispering, laughter and other shades of delivery
• Multilingual v2 speaks 29 languages, Flash v2.5 and Turbo v2.5 speak 32
• Turbo v2 is English-only

HeyGen Voice

HeyGen Voice and HeyGen Voice Turbo provide 10 ready-made voices. They are meant for video voice-overs and avatars.

Fish Audio Drama 3

Fish Audio Drama 3 is directed voice acting. You write the emotion in square brackets before a line, for example [nervous, whispering], and the model performs it that way. A two-voice dialogue comes out of a single request, with no stitching in pieces.

What left the catalog

The catalog is more compact now: outdated models are removed, and their pages lead to successors.

• GPT-5.6, all versions, to GPT-6.1 Sol and GPT-6 Luna
• GPT-5.5 to GPT-6 Astra
• Gemini 3.1 Pro to Gemini 4 Argon
• GPT-5.4 Nano to GPT-6 Luna
• Gemini 3.1 Flash Lite to Gemini 3.5 Flash Lite
• DeepSeek v3.2 to DeepSeek v4.1 Flash
• Grok 4 Fast to Grok 4.7
• Sora 2 and Sora 2 Pro to Veo 3.1 Fast and Veo 3.1
• Wan2.7 Video to Wan3.0 Video
• The Grok Imagine and Grok Imagine Pro image models to Grok Imagine Image 2.0

Old chats keep working: new messages in a chat with a removed model go to its replacement.

Which one to pick

Agentic development and business automation: GPT-6.1 Sol, and Sol Pro for hard cases where a mistake is costly. Features, bugs and polished documents: Claude Sonnet 5.5. Large streams of small tasks and subagents: Claude Haiku 5.5. Complex reasoning and long mixed-format documents: Gemini 4 Argon. Image editing and a product in a new scene: Nano Banana 2.1, and for running through variants quickly, Seedream 5.0 Flash. A fast low-cost clip: Grok Imagine Video 1.5 Lite or MiniMax H3 Fast. Multilingual voice-over with emotion tags: Eleven v4; ready voices for video: HeyGen Voice; an acted dialogue: Fish Audio Drama 3.

Everything here is in the catalog now and runs off the shared balance, with no separate subscription required.