ExpectedQ3 2026
Jamba 2
AI21's next hybrid Mamba-Transformer model. Expected to push further on long-context efficiency.
Try equivalent now: Gemini 2.5 Pro →
Claude Opus 5, GPT-6, Gemini 3 Ultra, Sora 2, Veo 4, DeepSeek V4 and 43+ more. Call the current best equivalent today — we auto-upgrade your routing the moment the new model ships. One API key for all 100+ current models plus every upcoming one.
AI21's next hybrid Mamba-Transformer model. Expected to push further on long-context efficiency.
Alibaba's next flagship Qwen model. Expected to remain among the strongest open-weight models, especially for Chinese and multilingual tasks.
Next multimodal Qwen with vision. Expected to be among the strongest open-weight vision-language models.
Anthropic's next flagship reasoning model, expected to extend Opus 4.x on long-horizon agentic tasks and coding. Will likely debut with a larger context window and improved tool-use reliability.
The next-gen ultra-fast, low-cost Claude tier. Expected to target sub-200ms first-token latency for high-volume agent traffic.
Rumored codename for an Anthropic creative-writing-tuned variant of Opus. Unconfirmed — tracked here so the page is live if/when it ships.
Next version of Anthropic's dedicated coding agent. Expected improvements in multi-file refactors and repo-wide navigation.
ByteDance's next Doubao Pro release. Dominant in Chinese consumer AI; increasingly relevant for multilingual traffic.
Cohere's next RAG-optimized flagship. Expected to extend grounding quality and citation reliability.
Next-gen Cohere multilingual embedding model after Embed v3. Expected to improve cross-lingual retrieval and introduce compressed int8 outputs.
Next generation of DeepSeek's general-purpose chat model. Expected to keep the disruptive price-performance ratio.
Successor to DeepSeek R1. Dedicated reasoning model expected to challenge OpenAI o-series at a fraction of the price.
Dedicated coding variant. Expected to target repo-wide navigation and agent workflows with extreme price efficiency.
Next-generation ElevenLabs voice cloning and TTS model. Expected to deliver higher fidelity multilingual cloning with lower latency than Multilingual v2.
Google's next frontier Gemini tier. Expected to extend the massive-context (1M+ tokens) advantage and close the reasoning gap vs. GPT and Claude.
Next-gen Gemini Flash. Expected to remain the cheapest-per-token frontier model on the market.
Google DeepMind's next text-to-video model, successor to Veo 3.x. Expected to extend duration and improve physics/motion fidelity.
Google's next-generation text-to-image model. Expected to push further on photorealism and text-rendering inside images.
Next-gen Ideogram image model known for best-in-class in-image typography. Expected to extend text rendering quality and add longer prompt adherence.
Next-generation Kuaishou Kling text-to-video model. Expected to extend clip length and improve physical realism beyond Kling 1.6.
Rumored Kuaishou Kling image generation model extending the Kling family beyond video. Expected to excel at photoreal Asian faces and product shots.
Next generation of the open-weight LLaVA vision-language model family.
Meta's next open-weight flagship. Expected to narrow the gap with frontier closed models while keeping fully open weights.
A refresh of Meta's largest instruction-tuned model. Expected to maintain the same 405B parameter count with a better post-training recipe.
Microsoft Research's next small-language-model release. Expected to extend the punch-above-weight reputation of Phi-3/4.
Next-generation MiniMax speech model after Speech 01 Turbo. Expected to improve zero-shot voice cloning and streaming TTS latency.
Next flagship from Mistral. Expected to close the gap with GPT-5 / Claude 4.5 while keeping the Mistral family's strong multilingual performance.
Mistral's next dedicated coding model. Expected to target fill-in-the-middle and fast local-style inference.
Next iteration of NVIDIA's Nemotron post-trained Llama models. Expected to keep chasing GPT-4-class quality in the 70B tier.
OpenAI's next-numbered flagship. Expected to extend GPT-5.x on reasoning, multimodality, and agentic tool use. Will likely launch in Pro, Standard, and Mini tiers.
The high-reasoning tier of GPT-5 for complex problem solving. Longer thinking budget, better math and code.
Next generation of OpenAI's dedicated reasoning models. Expected to beat o3 on math olympiad and coding benchmarks by a large margin.
Cheaper, faster o4 variant. Targets high-volume reasoning traffic where you don't need full o4 depth.
OpenAI's next text-to-video model. Extended duration, higher resolution, and better physical consistency over Sora 1.
Successor to DALL·E 3. Expected to push photorealism and fine-grained prompt adherence closer to Flux Pro / Midjourney v7.
Rumored next-gen OpenAI TTS model succeeding tts-1 / tts-1-hd. Expected to unify the Advanced Voice stack with an API-accessible synthesis endpoint.
Rumored next-gen OpenAI speech-to-text model after Whisper v3. Expected to push word error rates lower across long-tail languages and noisy audio.
Rumored successor to text-embedding-3-large/-small. Expected to push retrieval quality while keeping the same OpenAI embedding API.
Rumored next-gen Pika Labs video model after Pika 2.2. Expected to improve character consistency and motion coherence in longer clips.
Rumored Runway Gen-5 flagship video model after Gen-4. Expected to extend clip duration, improve camera control, and sharpen photoreal output.
Next numbered Stable Diffusion release. Expected to reset the open-weight image generation bar.
Successor to SDXL, the most widely used open-weight image base. Expected to target photorealism and anatomy fixes.
Stability AI's next-gen audio generation model after Stable Audio 2.5. Expected to extend track length and improve multi-instrument arrangements.
Next-generation Suno music generation model after v4.5. Expected to deliver longer coherent tracks with cleaner vocals and stems export.
Rumored next-gen Udio music model. Expected to improve vocal realism and multi-minute structure for full-song generation.
Next-generation Voyage AI retrieval embedding model after voyage-3. Expected to top MTEB leaderboards for code and long-context retrieval.
xAI's next flagship. Expected to widen real-time data access (X integration) and improve agentic reasoning.
Heavy-reasoning variant of Grok 4, similar to the Grok Heavy tier pattern.
xAI's dedicated coding model. Expected to compete directly with Claude Code and DeepSeek Coder on repo-wide tasks.