AI Model Comparison
57 active models compared side-by-side. LLMs, image generators, code models, embeddings, speech, and video, newest first. Retired models are listed separately at the bottom.
Last updated ·Pricing and lineup checked monthly, last checked
Sort by:
| Model | Provider | Params | Context | Input | Output | Type | Released |
|---|---|---|---|---|---|---|---|
|
Open
Closed
Multimodal
New
Key Strengths Description Source →Benchmarks
GPQA Diamond
SWE-bench Verified
AIME 2025
No benchmark data available for this model type. |
|||||||
· Released
Params:
Context:
In:
Open
Closed
Multimodal
New
No models match your search.
models shown
18 open-weight
37 multimodal
29 providers
Prices are list prices in USD per 1M tokens unless noted. Benchmarks are vendor-reported and shown only where published. Each model links to the source its data was checked against.
Retired & deprecated models 47
Models that have been shut down, deprecated, or replaced by a newer version from the same vendor, and what replaced them.
| Model | What happened |
|---|---|
| Claude Haiku 4.5Anthropic | Superseded by Claude Haiku 5.5, released 2026-10-07, which adds a 1M token context window at the same entry-level price tier. |
| Mistral Medium 3.5Mistral AI | Mistral deprecated Medium 3.5 effective 2026-10-31 (retiring 2027-04-30), recommending Mistral Large 4 as the replacement. |
| Nano Banana 2 (Gemini 3.1 Flash Image)Google | Superseded by Nano Banana 2.1, which Google released on 2026-10-06 as the direct successor. |
| GPT-4o mini TTSOpenAI | OpenAI marked GPT-4o mini TTS deprecated on its models page, with voice/transcription work shifting to the newer GPT-Realtime and GPT-Transcribe family. |
| Claude Sonnet 5Anthropic | Superseded by Claude Sonnet 5.5 (released Sept 28, 2026), which runs 30%+ faster at the same $2/$10 price; Sonnet 5 is now a legacy model on Anthropic's platform. |
| GPT-6 SolOpenAI | Replaced by GPT-6.1 Sol one week after launch, at the same list price but with a 95%-cheaper cached-input rate. |
| CodeLlamaMeta | Meta has released no code-specific Llama since 2024, and current open coding models far surpass it. |
| CodexOpenAI | The original 2021 Codex API models were shut down in 2023. The Codex name now refers to OpenAI's coding agent. |
| Cursor TabCursor | Cursor's in-editor autocomplete has no public specs. Cursor's listed model is now Composer 2.5. |
| DeepSeek Coder V2DeepSeek | DeepSeek no longer ships a separate coder line. Coding is built into the V4 models. |
| StarCoder 2BigCode | No newer release, and current open coding models far surpass it. |
| Cohere Embed v3Cohere | Superseded by Cohere Embed v4 (April 2025). |
| Voyage AIVoyage AI | The Voyage 3 generation was replaced by the Voyage 4 family (January 2026). |
| Claude 3 OpusAnthropic | Retired January 5, 2026. Replaced by Claude Opus 5.5. |
| Claude 3.5 SonnetAnthropic | Retired October 28, 2025. Replaced by Claude Sonnet 5. |
| Command R+Cohere | Deprecated by Cohere and superseded by Command A. |
| DBRXDatabricks | No successor from Databricks, and far outpaced by newer open models. |
| DeepSeek V3DeepSeek | Superseded by DeepSeek V4. Its API endpoints were retired July 24, 2026. |
| GPT-4 TurboOpenAI | Deprecated April 2026 with API shutdown on October 23, 2026. Replaced by the GPT-6 family. |
| GPT-4oOpenAI | Legacy in OpenAI's API and superseded by the GPT-6 family (Astra, Sol, Luna). |
| Gemini 1.5 FlashGoogle | Shut down September 24, 2025. Replaced by Gemini 3.8 Flash. |
| Gemini 1.5 ProGoogle | Shut down September 24, 2025. Replaced by Gemini 3.1 Pro. |
| Gemini UltraGoogle | The Gemini 1.0 Ultra tier was discontinued. Google's top tier is now Gemini Pro. |
| Granite 3.0IBM | Superseded by Granite 4.x, most recently Granite 4.2 (August 2026). |
| Grok-2xAI | Retired by xAI. Replaced by Grok 4.7. |
| Llama 3.1 405BMeta | Superseded by Llama 4 (April 2025). Weights remain downloadable. |
| Llama 3.1 70BMeta | Superseded by Llama 4 (April 2025). Weights remain downloadable. |
| Llama 3.1 8BMeta | Superseded by Llama 4 (April 2025). Weights remain downloadable. |
| Llama 4 MaverickMeta | Meta ended new Llama development in April 2026, shifting its own products to the proprietary Muse Spark model and releasing the open-weight Muse Glimmer 30B (August 2026) as its new open release. |
| Mistral 7BMistral AI | Retired from Mistral's API (weights still available). Replaced by the Ministral 3 series. |
| Mistral Large 2Mistral AI | Retired from Mistral's API (final version May 31, 2026). Replaced by Mistral Large 3 and Medium 3.5. |
| Mixtral 8x22BMistral AI | Retired from Mistral's API. Replaced by newer Mistral models. |
| Phi-3 MediumMicrosoft | Superseded by Microsoft's Phi-4 family. |
| Qwen 2.5 72BAlibaba | Superseded by several Qwen generations, most recently Qwen3.8. |
| DALL-E 3OpenAI | Removed from OpenAI's API on May 12, 2026. Replaced by GPT Image 2. |
| Flux.1Black Forest Labs | Superseded by FLUX.2 (November 2025). Still offered as a previous generation. |
| GPT Image 2OpenAI | Superseded by GPT Image 2.5 (Flare/Sunburst variants), released September 8, 2026, with higher image quality and about 50% lower generation latency. |
| Ideogram 2.0Ideogram | Two generations old. Replaced by Ideogram 4.0 (June 2026). |
| Midjourney v6Midjourney | Superseded by V7, then V8.1 (default since June 2026). |
| gpt-image-1OpenAI | Listed by OpenAI as its previous image model. Replaced by GPT Image 2. |
| ElevenLabsElevenLabs | Replaced in the table by ElevenLabs' current flagship model, Eleven v4. |
| Suno v4Suno | Retired when Suno v6 launched on September 9, 2026. |
| WhisperOpenAI | The whisper-1 API is deprecated (shutdown February 26, 2027). Replaced by GPT-Transcribe. Open weights remain available. |
| Pika 2.0Pika | Pika relaunched in September 2026 as an app hub running third-party models. |
| Runway Gen-3Runway | Gen-3 Alpha retired July 2026. Replaced by Gen-4.5. |
| SoraOpenAI | Sora 2 superseded the original, then the Sora app shut down April 26, 2026 and the API on September 24, 2026. |
| Stable Video DiffusionStability AI | No new general video model from Stability, and current video models far surpass it. |