> For the complete documentation index, see [llms.txt](https://docs.blockbrain.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.blockbrain.ai/news/llm-updates.md).

# LLM Updates

This section provides the latest updates to the LLMs available in Blockbrain. Users can learn about newly added models, model improvements, provider availability, retirements, and replacements.

{% updates format="full" %}
{% update date="2026-10-07" %}

## GPT 6.1 Sol, Claude Opus/Sonnet 5.5, Image 2.5 Flare, 

#### 1. GPT 6.1 Sol <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI & Azure (EU**</mark>*<mark style="color:$info;">)</mark>

Now available with EU hosting through OpenAI and Azure. A major efficiency upgrade over GPT 6 Sol, delivering near-Astra intelligence on agentic coding, computer use, and professional work.

The model is well suited for demanding scientific research, complex professional workflows, and high-volume agentic tasks that reuse context across multiple requests. Demonstrating a much stronger adherence to user intent and improved factual accuracy on difficult prompts. Best for teams that need top-tier coding and reasoning performance with its abilities to reduce factual errors on difficult prompts by roughly 32% at low reasoning effort.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 6 Sol

#### 2. Claude Opus 5.5

*<mark style="color:$info;">**Google (EU**</mark>*<mark style="color:$info;">)</mark>

Anthropic's most capable flagship model, now available with EU hosting through Google AI. It is built for complex, open-ended work requiring sustained judgment, deep reasoning, and careful execution.\
It is exceptionally strong for long-horizon professional workflows, deep research, and serving as the orchestrator for multi-step agentic systems. Best for teams tackling highly ambiguous or demanding tasks where maximum intelligence and strategic decision-making are more important than raw speed.reasoning effort.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Claude Opus 5

#### 3. Claude Sonnet 5.5

*<mark style="color:$info;">**AWS Bedrock & Google (EU**</mark>*<mark style="color:$info;">)</mark>

Now available with EU hosting through AWS Bedrock and Google AI. A clear upgrade over Sonnet 5, generating outputs over 30% faster while costing less for most tasks due to improved token efficiency. It brings a massive leap in agentic coding performance and is the first Sonnet model to launch with advanced cyber safeguards.

The model is well suited for well-scoped everyday knowledge work, rapid code iteration, bug fixing, and generating polished documents or spreadsheets. Best for teams that need a fast, highly capable collaborative partner for everyday work at an economical price.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Claude Sonnet 5

#### 4. GPT Image 2.5 Flare

*<mark style="color:$info;">**OpenAI (EU**</mark>*<mark style="color:$info;">)</mark>

Now available with EU hosting through OpenAI API. The new default standard for AI image generation, delivering sharper details, more precise editing, and higher fidelity at a 50% lower latency than GPT Image 2 while supporting resolutions up to 3840 × 2160 pixels (4K UHD). It excels at preserving original subjects from reference photos across multiple edits and supports complex layouts, including transparent backgrounds.

The model is particularly well suited for creator content, rapid UI prototyping, and high-volume generation workflows. Best for creative, marketing, and media teams needing fast, reliable, and production-ready visuals.

**Model type:** Image Generation\
**Max Resolution:** up to 4k resolution\
**Recommended upgrade from:** Claude Sonnet 5
{% endupdate %}

{% update date="2026-09-25" %}

## Anthropic Adds Text Watermarking on Sept 30

From September 30, 2026, Anthropic will add an EU AI Act text watermark to outputs from Claude Fable 5, Claude Sonnet 5 and Claude Opus 4.8. On AWS Bedrock, Google Cloud and Microsoft Foundry, rollout may take a few extra days.

The watermark is a subtle statistical pattern in word choice, applied by Anthropic to meet EU transparency requirements for AI-generated content. It adds no characters or tokens and does not affect meaning, quality, pricing, latency or request formats. It contains no user or organization data. No action is required.
{% endupdate %}

{% update date="2026-09-25" %}

## **Claude Opus 5.5, GPT 6 Sol/Luna and GPT Image 2.5 Sunburst**

#### 1. Claude Opus 5.5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

It is the successor to Claude Opus 5 and the first model in Anthropic's Claude 5.5 family, performing close to Claude Fable 5.1 on most work at a significantly lower cost. It leads on agentic coding, computer use, and knowledge work benchmarks, scoring 1846 Elo on GDPval-AA v2.1 (real-world professional tasks across 44 occupations) compared to 1735 for Fable 5.1 and 1542 for GPT 6 Astra, and 66.4% on Terminal-Bench 4.0 compared to 57.9% for GPT 6 Astra. It also writes more clearly than Opus 5, putting the most important information first and following style and formatting instructions more closely.

With a 1M token context window and up to 128K output tokens, it is well suited for large codebase migrations and code reviews, financial and legal analysis, research reports that require accurate figures and citations, and multi-step business workflows across connected tools. It completes tasks in fewer steps and with fewer tokens than Opus 5, which Anthropic estimates at about 40% lower cost on typical workloads, and generates output more than 30% faster. Anthropic applies stricter safeguards to cybersecurity and biology requests, so some tasks in these areas may be declined, while routine bug finding and fixing is unaffected. Best for teams that need near-flagship reasoning for complex, long-running work at a lower cost than Fable 5.1.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Claude Opus 5

#### 2. GPT 6 Sol <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI & Azure (EU)**</mark>*

Now available with EU hosting through OpenAI and Azure. It is the successor to GPT 5.6 Sol and the mid-tier model in OpenAI's GPT 6 family, trained with methods similar to GPT 6 Astra to bring much of Astra's gains in coding, factual accuracy, and computer use to a faster, lower-cost model. On OpenAI's internal factuality evaluation, it makes about half as many mistakes as GPT 5.6 Sol. On AutomationBench, which tests end-to-end business workflows across 47 tools, it scores 33.2%, outperforming GPT 6 Astra at low effort (30.3%) at roughly a quarter of the cost per task. It also communicates more clearly than its predecessor, with less jargon and slightly shorter answers.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 5.6 Sol

#### 3. GPT 6 Luna <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI & Azure (EU)**</mark>*

Now available with EU hosting through OpenAI and Azure. It is the successor to GPT 5.6 Luna and the most efficient model in OpenAI's GPT 6 family, built for focused, high-volume tasks where speed and cost matter more than maximum reasoning depth. Its performance is high for its price tier: at higher reasoning effort, it matches GPT 5.6 Sol on OpenAI's factuality evaluation at about 1% of the cost, and on AutomationBench it improves on GPT 5.6 Luna by 5.4 percentage points at 58% lower cost per task. It also scores 66.6% on DeepSWE, a benchmark of long-horizon software engineering tasks.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 5.6 Luna

#### 4. GPT Image 2.5 Sunburst <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT-image-2.5 Sunburst is now available with EU hosting through OpenAI. It is the successor to GPT-image-2 and OpenAI's most capable image model, built for workflows where editing precision matters most. It keeps subjects, faces, and compositions consistent across many rounds of revision, applies edits only where instructed, and renders sharp, legible text inside images. It holds the highest Answer Quality rating (4.8) of any image model in Blockbrain, and ranks #1 of 62 scored image models with an Artificial Analysis Arena Elo of 1196.

With output up to 3840px and support for up to 16 reference images in a single editing request, it is well suited for campaign creative, product imagery, branded assets, and slides or visuals with embedded text. It trades slightly longer generation times for extra detail. Best for teams producing client-facing visuals that need to hold up under close review.

**Model type:** Image Generation\
**Max Resolution:** Up to 4K\
**Recommended upgrade from:** GPT image 2
{% endupdate %}

{% update date="2026-09-08" %}

## **Gemini 3.8 Flash, GPT 6 Astra**

#### 1. Gemini 3.8 Flash <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google (EU)**</mark>*

Gemini 3.8 Flash is now available with EU hosting through Vertex AI. It is the successor to Gemini 3.7 Flash and Google's most intelligent workhorse model, delivering significant improvements in software engineering, agentic tasks, and multi-step reasoning. It is designed to work harder on complex tasks by executing extra reasoning steps and calling tools iteratively, all while maintaining the same speed and introductory low cost as 3.7 Flash.

With a 1M token context window, it is well suited for long-horizon coding, autonomous agents, and critical analysis in finance and law. Best for teams that want near-frontier reasoning and agent-first workflows without flagship pricing.

**Model type:** LLM Router\
**Context window:** 1M tokens\
**Recommended upgrade from**: Gemini 3.7 Flash

#### 2. GPT 6 Astra <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT 6 Astra is now available with EU hosting through OpenAI. It is OpenAI’s most intelligent and aligned model to date, setting a new standard for computer use, software engineering, and complex reasoning. It achieves human parity on novel problem-solving with a 99.9% score on ARC-AGI-3 while dramatically improving intent understanding and template adherence.

With a 1M token context window, it is well suited for autonomous data analysis, software testing, complex UI/UX generation, and navigating novel digital environments. It also introduces persistent context retrieval, allowing the model to remember critical details across long sessions without summarizing them away. Best for teams that require top-tier reasoning, strict template adherence, and the world's most capable computer-use model for complex multi-step automations.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 5.6 Sol
{% endupdate %}

{% update date="2026-08-25" %}

## **Auto Router (General) and Gemini 3.7 Flash**

#### 1. **Auto Router (General)** <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google (EU)**</mark>*

The Auto Router (General) is now available in production. It automatically routes each request to the right model based on task complexity, so users get quality answers without choosing a model themselves. Simple messages go to GPT 5.6 Luna, everyday and demanding work to Qwen 235B, and reasoning-heavy requests escalate to Claude Sonnet 5.

Across mixed workloads, the router returned answer quality comparable to a frontier model, at roughly 40% of the cost, with latency on par or faster. It is best used as a default for teams with varied workloads, where most requests do not need flagship capability but the option to escalate should stay open. Model routing is handled automatically and requires no configuration.

**Model type:** LLM Router\
**Context window:** 128k tokens\
**Recommended for:** Mixed workloads

#### 2. Gemini 3.7 Flash <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google (EU)**</mark>*

Gemini 3.7 Flash is now available with EU hosting through Vertex AI. It is the successor to Gemini 3.6 Flash and improves on coding, knowledge work, and web development. Google reports 65.3% on DeepSWE v1.1 (up from 49.0%) and 30.4% on AutomationBench for business workflows (up from 17.0%).

With a 1M token context window and text, image, audio, video, and PDF input, it is well suited for coding agents, multi-step automation, and everyday agentic workflows. Best for teams that want strong coding and tool-use performance without flagship pricing.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Gemini 3.6 Flash
{% endupdate %}

{% update date="2026-08-18" %}

## **Gemini 3.6 Flash, Kimi K3, Qwen3 Next 80B A3B Thinking and GPT OSS 120B**

#### 1. Gemini 3.6 Flash <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google (EU)**</mark>*

Gemini 3.6 Flash is now available with **EU hosting through Vertex AI**. It is the successor to Gemini 3.5 Flash and Google's workhorse model, delivering better coding, knowledge work, and multimodal quality while meaningfully improving token efficiency. Google reports roughly 17% fewer output tokens than 3.5 Flash on the Artificial Analysis Index.

With a **1M token context window** and text, image, audio, video, and PDF input, it is well suited for agentic workflows, long document analysis, chart and visual reasoning, and everyday coding. It also completes multi-step workflows in fewer turns and shows lower compile-failure and revision rates on code generation. Best for teams that want one model covering multi-step tool use and multimodal work without flagship pricing.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Gemini 3.5 Flash

#### 2. Kimi K3 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Nebius (EU)**</mark>*

Kimi K3 is now available with **EU hosting through Nebius**. It is Moonshot AI's flagship open-weight model and the first open-source model in the 3-trillion-parameter class at 2.8T parameters, with an Artificial Analysis Intelligence Index score of 57.1, the strongest open-weight result currently tracked.

With a **1M token context window**, native vision and always-on reasoning, it is built for long-horizon coding, large-repository work, agentic research, and document and visual analysis. It scored 93.5% on GPQA Diamond and 91.2% on BrowseComp at launch. Best for demanding workloads where teams want frontier-adjacent quality on an open-weight model hosted in the EU.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Kimi K2.5

#### 3. Qwen3 235B A22B Instruct <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Nebius (EU)**</mark>*

Qwen3 235B A22B Instruct is now available with **EU hosting through Nebius**. It is a Mixture-of-Experts model with 235B total parameters and only 22B active per token, which keeps inference cost low relative to its capability. It is not a current-generation frontier release, but it remains a dependable general-purpose option.

With a **262K token context window**, it handles multilingual work, instruction following, structured output, and tool use well. Best for high-volume general workloads where cost matters more than top-tier reasoning, and for teams that need an open-weight fallback under EU hosting.

**Model type:** LLM\
**Context window:** 262K tokens

#### 4. GPT OSS 120B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Nebius (EU)**</mark>*

GPT OSS 120B is OpenAI's open-weight Mixture-of-Experts model, now available with **EU hosting through Nebius**. It has 117B total parameters and activates 5.1B per forward pass, and supports configurable reasoning depth, full chain-of-thought access, and native tool use including function calling and structured output.

With a **131K token context window**, it is suited to high-throughput reasoning tasks, classification, extraction, and tool-calling agents where latency and cost are the constraint. Nebius list pricing is $0.15 per 1M input and $0.60 per 1M output tokens. Best for teams that need cheap, fast open-weight inference in the EU with adjustable reasoning effort.

**Model type:** LLM\
**Context window:** 131K tokens
{% endupdate %}

{% update date="2026-08-05" %}

## **GPT 5.6 Fast Mode and Gemini 3.5 Flash (Lite)**

#### 1. GPT 5.6 Sol (Fast) <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT 5.6 Sol (Fast) is now available with **EU hosting through OpenAI**. It delivers the flagship intelligence of GPT 5.6 Sol at up to **2.5× faster speeds**, making it a strong choice when both maximum capability and low latency are essential.

With a **1M** token context window, It is especially well suited for complex coding, advanced reasoning, long-running agents, research, and other demanding workflows where teams need top-tier results without waiting for standard flagship response times. Fast Mode retains the same intelligence as GPT 5.6 Sol at approximately twice the price.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 5.6 Sol when faster responses are required

#### 2. GPT 5.6 Terra (Fast) <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT 5.6 Terra (Fast) is now available with **EU hosting through OpenAI**. It provides the same balanced intelligence as GPT 5.6 Terra with accelerated response times, offering near-flagship quality for everyday workflows where latency matters.

With a **1M** token context window, it is well suited for time-sensitive coding, document analysis, tool use, knowledge work, and multi-step automation. It is a practical middle ground for teams that need stronger capability than lightweight models but do not require the full depth or cost of the Sol tier. Fast Mode is priced at approximately twice the standard version.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 5.6 Terra when faster responses are required

#### 3. GPT 5.6 Luna (Fast) <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT 5.6 Luna (Fast) is now available with **EU hosting through OpenAI**. Built for maximum throughput, it delivers the same intelligence as GPT 5.6 Luna with accelerated response speeds, making it ideal for workloads where fast and consistent output is the priority.

With a **1M** token context window, it is especially well suited for high-volume automations, data extraction, classification, customer-facing assistants, document processing, and other time-critical production tasks. It retains Luna’s strong cost efficiency while providing near-instant responses, with Fast Mode priced at approximately twice the standard version.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** GPT 5.6 Luna when faster responses are required

#### 4. Gemini 3.5 Flash (Lite) <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

Gemini 3.5 Flash-Lite is now available with **EU hosting through Google AI**. It is the fastest and most cost-effective model in the Gemini 3.5 family, designed for speed, scale, and high-volume production workloads.

With a **1M** token context window and output speeds of up to 350 tokens per second, it is a major upgrade over Gemini 3.1 Flash-Lite. It is particularly well suited for agentic search, document processing, data extraction, classification, and other high-throughput workflows where minimizing latency and cost is more important than using a larger general-purpose model.

**Model type:** LLM\
**Context window:** 1M tokens\
**Recommended upgrade from:** Gemini 3.1 Flash (Lite)
{% endupdate %}

{% update date="2026-07-31" %}

## **Qwen, Llama, GPT and Gemma**

#### 1. Qwen 3VL 235B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**StackIT (EU)**</mark>*

Qwen 3VL 235B is now available with **EU hosting through StackIt**. It is a powerful open-weight vision-language model designed for advanced multimodal work, including long-document processing, extended video understanding, complex visual analysis, and agentic workflows.

With a **218K** token context window, it is especially well suited for teams that need high-end multimodal capabilities and have the infrastructure to support a heavyweight model.

**Model type:** LLM\
**Context window:** 218K tokens

#### 2. Qwen3.6 27B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**StackIT (EU)**</mark>*

Qwen3.6 27B is now available with **EU hosting through StackIt**. It is an open-weight multimodal model that delivers particularly strong agentic coding capabilities for its size. The model is well suited for software development, frontend workflows, tool use, visual analysis, and other tasks that combine coding with multimodal understanding.

With a **262K** token context window, it offers a strong balance of capability, deployment flexibility, and long-context support.

**Model type:** LLM\
**Context window:** 262K tokens

#### 3. Llama 3.3 70B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**StackIT (EU)**</mark>*

Llama 3.3 70B is now available with **EU hosting through StackIt**. It is a balanced, high-performance open-weight model that provides robust reasoning and consistently reliable output quality across a broad range of tasks.&#x20;

With a **128K** token context window, it is especially well suited for complex analysis, long-form content creation, contextual decision support, and general-purpose enterprise workflows that require dependable performance.

**Model type:** LLM\
**Context window:** 128K tokens

#### 4. GPT OSS 120B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**StackIT (EU)**</mark>*

GPT OSS 120B is now available with **EU hosting through StackIt**. It is OpenAI’s largest open-weight reasoning model, built for advanced coding, reasoning, tool calling, and agentic workflows.&#x20;

With a **131K** token context window, it is especially well suited for enterprise agents, complex development tasks, and specialized deployments where organizations require greater control over their data and model infrastructure.

**Model type:** LLM\
**Context window:** 131K tokens

#### 5. GPT OSS 20B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**StackIT (EU)**</mark>*

GPT OSS 20B is now available with **EU hosting through StackIt**. It is a compact open-weight reasoning model designed for low-latency, private, and specialized deployments. The model supports coding, mathematics, tool use, and agentic workflows while requiring as little as 16 GB of memory.&#x20;

With a **131K** token context window, it is especially well suited for private agents, local inference, and cost-conscious deployments that need capable reasoning without heavyweight infrastructure.

**Model type:** LLM\
**Context window:** 131K tokens

#### 6. Gemma 3 27B <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**StackIT (EU)**</mark>*

Gemma 3 27B is now available with **EU hosting through StackIt**. It is a capable open-weight multimodal model supporting text and image inputs, more than 140 languages, function calling, and structured outputs.

With a **131K** token context window, it is especially well suited for document analysis, multilingual assistants, visual tasks, and flexible agentic workflows that require reliable multimodal understanding in a comparatively compact model.

**Model type:** LLM\
**Context window:** 131K tokens
{% endupdate %}

{% update date="2026-07-28" %}

## **Claude Opus 5**

#### 1. Claude Opus 5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Opus 5 is now available with **EU hosting through AWS Bedrock**. It is a major step up from Opus 4.8 at the same price, delivering near–Fable 5 intelligence at approximately half the cost. The model is especially strong at complex professional work and long-running agent workflows where it must review, verify, and improve its own output.&#x20;

With a 1M token context window, Opus 5 is the recommended upgrade for AWS Bedrock users who need greater intelligence and reliability without an increase in model pricing.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Claude Opus 4.8

#### 2. Claude Opus 5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google AI (EU)**</mark>*

Claude Opus 5 is now available with **EU hosting through Google AI.** It is a major step up from Opus 4.8 at the same price, delivering near–Fable 5 intelligence at approximately half the cost. The model is especially strong at complex professional work and long-running agent workflows where it must review, verify, and improve its own output.&#x20;

With a 1M token context window, Opus 5 is the recommended upgrade for AWS Bedrock users who need greater intelligence and reliability without an increase in model pricing.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Claude Opus 4.8
{% endupdate %}

{% update date="2026-07-23" %}

## Grok 4.5

#### 1. Grok 4.5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**xAI (US)**</mark>*

**Grok 4.5** is now available with **US hosting through xAI**. xAI’s latest flagship model combines strong reasoning with fast response speeds and competitive pricing, with particular strengths in real-world software engineering, coding agents, and agentic task execution.&#x20;

With a **500k** token context window, it is especially well suited for technical workflows, application development, business productivity, and complex knowledge work that requires high capability without the latency and cost of slower flagship models.

**Model type:** LLM\
**Context window:** 500,000 tokens\
**Recommended upgrade from:** Grok 4
{% endupdate %}

{% update date="2026-07-16" %}

## **Claude Sonnet 5**

#### 1. Claude Sonnet 5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google AI (EU)**</mark>*

Claude Sonnet 5 is now available with **EU hosting through Vertex AI**. It is a major upgrade over Sonnet 4.6, delivering near-flagship performance across reasoning, coding, tool use, and knowledge work at a lower cost than top-tier models.&#x20;

With a **1M** token context window, Sonnet 5 is especially well suited for complex, multi-step tasks, large document and codebase analysis, and automation workflows that require high capability without premium pricing.&#x20;

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Claude Sonnet 4.6&#x20;
{% endupdate %}

{% update date="2026-07-15" %}

## **Gemini 3.5 Flash, Mistral Medium 3.5 and Claude Models**

#### 1. Gemini 3.5 Flash <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google AI (EU)**</mark>*

Gemini 3.5 Flash is now available with **EU hosting through Google AI**. It is a major step up from Gemini 3.1 Pro, combining strong agentic, coding, financial, and multimodal capabilities with the speed expected from a Flash model.

Its **1M** token context window makes it particularly effective for fast multi-step task handling, coding workflows, document analysis, and automation.&#x20;

The model offers **Minimal, Low, Medium, and High** thinking modes, allowing users to balance response speed with deeper reasoning depending on the task.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Gemini 2.5 Flash

#### 2. Mistral Medium 3.5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Mistral (EU)**</mark>*

Mistral Medium 3.5 is now available with **EU hosting through Mistral**. This open-weight model combines Mistral’s previous reasoning and coding lines into a unified model, providing strong performance for agentic coding, tool use, and multi-step technical work.&#x20;

With a **256k** token context window, it is particularly well suited for coding agents, software development workflows, technical automation, and applications that need a capable general-purpose model with open-weight flexibility.

**Model type:** LLM\
**Context window:** 256,000 tokens\
**Recommended upgrade from:** Mistral Medium

#### 3. Claude Opus 4.8 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Opus 4.8 is now available with **EU hosting through AWS Bedrock**. It is a significant upgrade over Opus 4.7, with stronger capabilities across agentic coding, computer use, and complex long-running tasks. Improved honesty and reliability also make it a more dependable partner for workflows that require sustained autonomy and accurate reporting.&#x20;

With a **1M** token context window, it is especially well suited for difficult engineering work, computer-use agents, codebase management, and high-stakes knowledge tasks.&#x20;

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Claude Opus 4.7reyouki

#### 4. Claude Opus 4.7 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Opus 4.7 is now available with **EU hosting through AWS Bedrock**. It is a major upgrade over Opus 4.6, designed for difficult hands-on coding and engineering work that can be delegated with greater confidence. Beyond coding, it provides stronger performance across reasoning, tool use, and complex knowledge tasks.&#x20;

With a **1M** token context window, it is particularly well suited for advanced software development, codebase analysis, long-running agents, and demanding multi-step workflows. For AWS Bedrock users of Opus 4.6, Opus 4.7 is the recommended upgrade.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Claude Opus 4.6

#### 5. Claude Opus 4.6 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Opus 4.6 is now available with **EU hosting through AWS Bedrock**. It improves on Opus 4.5 across coding, agentic task execution, codebase management, structured reviews, and debugging. It also provides more consistent and efficient performance during demanding analysis, document review, and multitasking workflows.&#x20;

With a **1M** token context window, it is well suited for complex engineering, long-running agents, large-scale document work, and tasks requiring reliable reasoning across multiple steps.

The model offers **Low, Medium, High, and Max Reasoning** modes, ranging from faster and more efficient responses to maximum reasoning depth for the most complex tasks.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Thinking modes:** Low Reasoning, Medium Reasoning, High Reasoning, Max Reasoning\
**Recommended upgrade from:** Claude Opus 4.5

#### 6. Claude Sonnet 4.6 (Fast) <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Sonnet 4.6 (Fast) is now available with **EU hosting through AWS Bedrock**. This configuration operates strictly in **Non-Reasoning mode**, bypassing extended thinking steps to deliver lower-latency responses while retaining the underlying quality of Sonnet 4.6.&#x20;

With a **1M** context window, it is best suited for responsive chat experiences, high-volume production workloads, straightforward coding assistance, document processing, and agent workflows where speed is more important than deeper reasoning.

**Model type:** LLM\
**Context window:** 1 million tokens

#### 7. Claude Haiku 4.5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Haiku 4.5 is now available with **EU hosting through AWS Bedrock**. It is a fast and cost-efficient model that delivers strong coding, reasoning, tool use, and interface interaction capabilities at production scale. Its speed and ability to execute tasks in parallel make it particularly effective as a worker model within multi-agent systems.

With a **200k** token context window and an available Fast configuration, it is well suited for backend automation, chat workloads, coding assistance, and high-volume agent systems that prioritize responsiveness and lower costs.

**Model type:** LLM\
**Context window:** 200,000 tokens\
**Recommended upgrade from:** Claude Haiku 3.5
{% endupdate %}

{% update date="2026-07-14" %}

## **Claude Sonnet 5, Gemini 3.1 Flash (Lite) Image, GPT 5.6 Models**

#### 1. Claude Sonnet 5 <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**AWS Bedrock (EU)**</mark>*

Claude Sonnet 5 is now available with **EU hosting through AWS Bedrock**. It is a major upgrade over Sonnet 4.6, delivering near-flagship performance across reasoning, coding, tool use, and knowledge work at a lower cost than top-tier models.

With a **1M** token context window, Sonnet 5 is especially well suited for complex, multi-step tasks, large document and codebase analysis, and automation workflows that require high capability without premium pricing.&#x20;

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** Claude Sonnet 4.6

#### 2. Gemini 3.1 Flash-Lite Image <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Google AI (US)**</mark>*

Gemini 3.1 Flash-Lite Image is now available with **US hosting through Google AI**. This lightweight image-generation model provides a fast and cost-efficient option for creating visuals directly within the platform. It is particularly well suited for high-volume image generation, rapid creative experimentation, simple marketing assets, and workflows where turnaround time and affordability are more important than the advanced controls of larger image models.

**Model type:** Image model\
**Provider:** Google AI\
**Hosting:** US

#### 3. GPT 5.6 Sol <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Azure AI (EU)**</mark>*

GPT-5.6 Sol is now available with **EU hosting through Azure AI**. It is the most capable model in the GPT-5.6 family, designed to deliver the highest-quality results for complex and demanding work.&#x20;

With a **1M** token context window, it is especially well suited for difficult reasoning, long-running tasks, advanced research, complex coding, document-intensive analysis, and workflows where accuracy and depth are more important than speed or cost.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** GPT 5.5

#### 4. GPT 5.6 Terra <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Azure AI (EU)**</mark>*

GPT-5.6 Sol is now available with **EU hosting through Azure AI**. It is the most capable model in the GPT-5.6 family, designed to deliver the highest-quality results for complex and demanding work.&#x20;

With a **1M** token context window, it is especially well suited for difficult reasoning, long-running tasks, advanced research, complex coding, document-intensive analysis, and workflows where accuracy and depth are more important than speed or cost.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** GPT 5.4

#### 5. GPT 5.6 Luna <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**Azure AI (EU)**</mark>*

GPT-5.6 Sol is now available with **EU hosting through Azure AI**. It is the most capable model in the GPT-5.6 family, designed to deliver the highest-quality results for complex and demanding work.&#x20;

With a **1M** token context window, it is especially well suited for difficult reasoning, long-running tasks, advanced research, complex coding, document-intensive analysis, and workflows where accuracy and depth are more important than speed or cost.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** GPT 5.4, GPT 5.4 Nano, GPT 5.4 Mini
{% endupdate %}

{% update date="2026-07-10" %}

## **GPT 5.6 Models**

#### 1. GPT 5.6 Sol <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT-5.6 Sol is now available with **EU hosting through OpenAI**. It is the most capable model in the GPT-5.6 family, designed to deliver the highest-quality results for complex and demanding work.&#x20;

With a **1M** token context window, it is especially well suited for difficult reasoning, long-running tasks, advanced research, complex coding, document-intensive analysis, and workflows where accuracy and depth are more important than speed or cost.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** GPT 5.5

#### 2. GPT 5.6 Terra <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT-5.6 Sol is now available with **EU hosting through OpenAI**. It is the most capable model in the GPT-5.6 family, designed to deliver the highest-quality results for complex and demanding work.&#x20;

With a **1M** token context window, it is especially well suited for difficult reasoning, long-running tasks, advanced research, complex coding, document-intensive analysis, and workflows where accuracy and depth are more important than speed or cost.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** GPT 5.4

#### 3. GPT 5.6 Luna <a href="#sharepoint-page-reading-for-agents" id="sharepoint-page-reading-for-agents"></a>

*<mark style="color:$info;">**OpenAI (EU)**</mark>*

GPT-5.6 Sol is now available with **EU hosting through OpenAI**. It is the most capable model in the GPT-5.6 family, designed to deliver the highest-quality results for complex and demanding work.&#x20;

With a **1M** token context window, it is especially well suited for difficult reasoning, long-running tasks, advanced research, complex coding, document-intensive analysis, and workflows where accuracy and depth are more important than speed or cost.

**Model type:** LLM\
**Context window:** 1 million tokens\
**Recommended upgrade from:** GPT 5.4, GPT 5.4 Nano, GPT 5.4 Mini
{% endupdate %}
{% endupdates %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.blockbrain.ai/news/llm-updates.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.