Blog · Jun 7, 2026

5 New AI Models Added in June 2026: Pricing, Context Windows & When to Use Each

The LLM landscape just got 5 new options — from DeepSeek V3.2 at $0.23/M input to Cohere Command A at $2.50/M. Here's what each one costs and when to use it.

June 2026 brought 5 new models to the APIpulse pricing database, bringing our total to 39 models across 10 providers. Whether you're building a chatbot, running RAG pipelines, or optimizing costs at scale, there's likely a new option that fits your budget better than what you're using today.

The 5 New Models at a Glance

All prices per 1M tokens. Verified Jun 7, 2026.

1. DeepSeek V3.2 — The Budget King

At $0.23/$0.34 per 1M tokens, DeepSeek V3.2 is the cheapest new model by a wide margin. It's 6x cheaper on input than the next cheapest new model (Gemini 3.5 Flash at $1.50) and 26x cheaper on output.

Best for: High-volume classification, content moderation, chatbot backends, code autocomplete — anything where you're processing millions of tokens and cost is the primary concern.

Trade-off: 128K context (not 1M). If you need to process entire codebases or long documents in a single prompt, you'll need a model with a larger window.

2. Gemini 3.5 Flash — Google's Speed Demon

Google's latest Flash model at $1.50/$9.00 per 1M tokens with a massive 1M token context window. It's designed for speed — fast inference, low latency, and native multimodal support.

Best for: Google Cloud workloads, enterprise applications needing SOC 2/HIPAA compliance, multimodal tasks (image + text), and long-document processing.

Trade-off: Output pricing at $9.00/M is steep. If your workload is output-heavy (long responses, content generation), costs add up fast.

3. Mistral Medium 3.5 — Europe's Mid-Tier Contender

Mistral's Medium 3.5 at $1.50/$7.50 per 1M tokens is priced identically to Gemini 3.5 Flash on input but 17% cheaper on output. The key differentiator: European data sovereignty.

Best for: EU-based companies with GDPR requirements, applications needing European data residency, and teams that want a mid-tier model without US cloud lock-in.

Trade-off: 128K context vs Gemini's 1M. If context window size matters, Gemini wins.

4. AI21 Jamba 1.7 Large — The Hybrid Architecture

AI21's Jamba 1.7 Large at $2.00/$8.00 per 1M tokens uses a hybrid Transformer-Mamba architecture, which means faster inference on long sequences. The 256K context is a solid middle ground.

Best for: Long-document summarization, enterprise RAG pipelines, and applications where inference speed on long contexts matters more than raw cost.

Trade-off: More expensive than DeepSeek and Mistral. The hybrid architecture shines on long sequences but may not be cost-competitive for short prompts.

5. Cohere Command A — Enterprise RAG Specialist

Cohere's Command A at $2.50/$10.00 per 1M tokens is the most expensive new model, but it's optimized for enterprise RAG and retrieval-heavy workloads. Cohere's embedding + generation ecosystem is tightly integrated.

Best for: Enterprise RAG pipelines using Cohere Embed, applications needing strong grounding and citation, and teams already in the Cohere ecosystem.

Trade-off: Highest cost of the 5 new models. Only makes sense if you're leveraging Cohere's full stack (embed + rerank + generate).

How These Compare to Existing Models

Here's how the new models stack up against the popular incumbents:

• DeepSeek V3.2 vs GPT-5: 97% cheaper on input ($0.23 vs $7.50). GPT-5 has better reasoning, but for classification/summarization, DeepSeek is a no-brainer.
• Gemini 3.5 Flash vs Claude Sonnet 4.6: 50% cheaper on input ($1.50 vs $3.00), 40% cheaper on output ($9.00 vs $15.00). Gemini has 1M context vs Claude's 1M — similar specs, lower price.
• Mistral Medium 3.5 vs Claude Sonnet 4.6: Same input price ($1.50 vs $3.00 — actually 50% cheaper), 50% cheaper output. European alternative with comparable capabilities.
• AI21 Jamba 1.7 vs Gemini 2.5 Pro: Similar pricing ($2.00 vs $1.25 input), but Jamba's hybrid architecture may be faster on long contexts.
• Cohere Command A vs GPT-5: 67% cheaper on input ($2.50 vs $7.50). Best for teams already using Cohere's embedding/reranking stack.

Which One Should You Use?

The Bigger Picture

These 5 models push the APIpulse database to 39 models across 10 providers. The trend is clear: prices keep falling, context windows keep growing, and developers have more choices than ever.

The best strategy in 2026 isn't picking one model — it's using a multi-model approach. Use DeepSeek V3.2 for high-volume cheap tasks, Gemini 3.5 Flash for long-context work, and GPT-5 or Claude Opus 4.8 for complex reasoning. Our decision tree can help you pick.