July 2026 Model Updates | CipherSense Agents

July has been one of the busiest months we've seen for frontier models. In the space of three weeks, Anthropic, OpenAI, Google, xAI, and Moonshot AI all shipped new flagships — several of them arriving within days of each other. We've rolled all of it into CipherSense Agents' model picker as it landed, so every workflow, agent node, and API-connected provider on the platform can use these models today.

Here's a rundown of what's new, provider by provider, and a few notes on when to reach for each one.

Anthropic: Claude Opus 5, Fable 5, Opus 4.8, and Sonnet 5

Anthropic shipped four new models this cycle, refreshing every tier of the Claude lineup — most recently Claude Opus 5, which landed just as we were finishing this post:

Claude Opus 5 (claude-opus-5) — the new flagship of the Opus tier, succeeding Opus 4.8. A step-change on deep reasoning, agentic coding, and long-horizon autonomous work, at the same $5/$25 per-million-token pricing as its predecessor. This is now our recommended Opus-tier pick on the platform. 1M token context window.
Claude Fable 5 (claude-fable-5) — Anthropic's most capable widely-released model. Thinking is always on, and it's built for the most demanding reasoning and long-horizon agentic work — the kind of run that used to need constant human check-ins. 1M token context window.
Claude Opus 4.8 (claude-opus-4-8) — the previous top of the Opus tier. State-of-the-art on long-horizon agentic execution, knowledge work, and memory, with a noticeably warmer, clearer writing style than its predecessor. Still fully supported. 1M token context window.
Claude Sonnet 5 (claude-sonnet-5) — near-Opus quality on coding and agentic work at Sonnet-tier cost. This remains our recommended default for Anthropic across the platform.

If you were running production workflows on Claude Sonnet 4.6, Opus 4.7, or Opus 4.8, all three are still available and fully supported — nothing breaks. But for new agent nodes: start with Sonnet 5 for most workflows, reach for Opus 5 when a task genuinely needs Opus-tier depth, and go to Fable 5 only for the hardest, longest-horizon problems.

OpenAI: the GPT-5.6 family — Sol, Terra, Luna

OpenAI moved to a three-tier model family with GPT-5.6, replacing the single-flagship approach with models tuned for different points on the cost/latency/quality curve:

GPT-5.6 Sol (gpt-5.6-sol) — the flagship. Deepest reasoning, positioned for the most complex agentic and coding workloads.
GPT-5.6 Terra (gpt-5.6-terra) — the balanced, production-default tier. Performance competitive with GPT-5.5 at roughly half the price of Sol — this is now CipherSense's default OpenAI model.
GPT-5.6 Luna (gpt-5.6-luna) — the fast, cheap tier for high-volume work where throughput matters more than squeezing out the last few points of quality.

All three share a 1.05M token context window and 128K max output, so switching between tiers mid-workflow doesn't mean re-thinking your prompt structure — just your cost/latency budget.

Google Gemini: 3.6 Flash and 3.5 Flash-Lite

Google skipped straight past a 3.5 Pro release (still in partner testing) and shipped two new models in the Flash line instead:

Gemini 3.6 Flash (gemini-3.6-flash) — Google's new workhorse model. Stronger on complex agentic and multimodal tasks than 3.5 Flash, while using fewer tokens per task — which in practice means it's both better and cheaper to run. Now our default Gemini model.
Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite) — the most cost-effective model in the current Gemini class, built for high-volume agents and document-processing pipelines where every token counts.

Both carry Gemini's usual 1M token context window and full multimodal input support (text, image, video, audio, PDF).

xAI: Grok 4.5

xAI's new flagship, Grok 4.5 (grok-4.5), built on their V9 foundation model, is squarely focused on coding, agentic workflows, and knowledge work, with configurable reasoning effort. It's a 500K-context model that xAI positions as competitive with Claude Opus 4.7 — at a noticeably lower price and faster inference. It's now the recommended default for Grok on the platform, alongside the existing multi-agent and fast-reasoning variants.

Moonshot AI: Kimi K3 (via Ollama Cloud)

Kimi is accessed through our Ollama (Cloud) integration, and it just got a serious upgrade. Kimi K3 (kimi-k3) is Moonshot AI's new flagship: a 2.8-trillion-parameter model with native visual understanding, a 1M token context window, and an always-on "thinking mode." Independent benchmarks have it ranking ahead of Claude Opus 4.8 on real-world task evaluations — a notable result for an open-weight model.

We've also added Kimi K2.6 (kimi-k2.6) alongside it, which introduced the Agent Swarm system (scaling to 300 coordinated sub-agents) that K3 builds on. Both sit next to the existing Kimi K2 and K2.5 variants in the Ollama model list, so nothing you've already built on Kimi needs to change.

Also new: Llama 4, Qwen 3.6, and Mistral Small 4 on HuggingFace

Rounding out this update, three new open-weight models are now available through our HuggingFace integration:

Llama 4 Maverick (meta-llama/Llama-4-Maverick-17B-128E-Instruct) — Meta's natively multimodal Mixture-of-Experts model, now our HuggingFace default.
Llama 4 Scout (meta-llama/Llama-4-Scout-17B-16E-Instruct) — same family, tuned for exceptionally long context (up to 10M tokens) — useful if you need to reason over an entire codebase or document set in one pass.
Qwen 3.6-27B (Qwen/Qwen3.6-27B) — Alibaba's latest open-weight release, punching well above its parameter count on coding benchmarks, with an extendable context window up to 1M tokens.
Mistral Small 4 (mistralai/Mistral-Small-4-119B-2603) — Mistral's unified instruct, reasoning, and agentic-coding model (folding what used to be three separate lines — Magistral, Pixtral, and Devstral — into one), with native vision input.

How to switch

Nothing about your existing workflows changes automatically — models you've already selected keep running exactly as configured. To try one of the new additions:

Go to Organization Settings → LLM Providers and confirm the relevant provider is connected.
Open any agent node in the workflow designer and pick the new model from the dropdown, or reference it by ID (e.g. @claude-sonnet-5, @gpt-5.6-terra) in the AI designer chat.
For providers where we've updated the recommended default (Anthropic, OpenAI, Google, xAI, HuggingFace), new nodes will suggest the latest model automatically — you can always override it per node.

We'll keep this cadence going: as providers ship new frontier models, we aim to have them in the picker within days, not months. If there's a model you're waiting on that isn't listed yet, let us know.

Questions about which model fits your workflow? Explore the workflow canvas or check the AI agent provider docs for full specs, context windows, and setup guides for every supported provider.