{"course":{"slug":"choosing-your-tools","title":"Choosing and Using the Tools","url":"https://addaly.com/learn/choosing-your-tools","level":"beginner","track":"work"},"lesson":{"slug":"reading-a-model-name","title":"Reading a model's name","url":"https://addaly.com/learn/choosing-your-tools/reading-a-model-name","position":2,"of":76,"minutes":9,"answer":"A model name encodes family, version, size class and build date, and the size class and the build date are the two parts that quietly decide what you actually get.","keyIdea":"A model name encodes family, version, size class and build date, and the size class and the build date are the two parts that quietly decide what you actually get."},"body":{"format":"markdown","text":"## The name is a specification\n\n`claude-3-5-sonnet-20241022`. `gpt-4o-mini`. `gemini-2.5-flash`. `llama-3.1-70b-instruct`. `qwen3-8b`. `deepseek-r1`. These look like noise until you know the parts, and then each one tells you roughly what you are dealing with before you send a single message.\n\nAlmost every name carries four things, in some order.\n\n**The family.** Who made it and which lineage it belongs to. GPT, Claude, Gemini, Llama, Mistral, Qwen, DeepSeek, Gemma, Phi.\n\n**The version.** A number that goes up: 3, 3.5, 4, 4.1. A bigger number from the same maker is usually a better model, and this is the one part of the name you can mostly trust.\n\n**The size or speed label.** This is the part people miss. `mini`, `nano`, `flash`, `lite`, `small`, `haiku` mean the cheap fast one. `pro`, `opus`, `large`, `ultra` mean the expensive slow one. `sonnet`, `medium`, and unlabelled names usually sit in the middle. Open models put the parameter count in directly: `8b` is eight billion parameters, `70b` is seventy.\n\n**The build or date.** `20241022` is a snapshot from 22 October 2024. `-preview`, `-exp`, `-latest` are channels rather than dates, and they behave differently, which the last lesson of this module is about.\n\nTwo extra words appear on downloadable models. `instruct` or `chat` means it has been trained to follow instructions and hold a conversation: this is the one you want. A **base** model with no such suffix continues text rather than answering it; ask it a question and it may write three more questions. People download the base model by mistake, conclude the model is broken, and give up.\n\n## Version numbers are not comparable across companies\n\nGemini 2.5 is not \"worse than\" GPT-4 because 2.5 is less than 4. The numbers are internal to each company and reset whenever marketing decides. OpenAI has shipped `gpt-4o` and `o4-mini` as different products in the same period; one is a general model, the other from a reasoning line. Read the family first, the number second, and never compare two makers' version numbers directly.\n\n## Finding out what is actually answering you\n\nIn a chat product, look in three places, in this order:\n\n1. The model picker at the top of the conversation. It names a product tier, not always the exact model.\n2. Settings, then the account or model page.\n3. The small print under an individual answer: some products name the model there, especially after a fallback.\n\nIf it says something like *\"Fast\"*, *\"Smart\"*, or *\"Auto\"*, you have been given a routing label rather than a model name. That is a real answer too: it means the product is choosing for you and may choose differently tomorrow.\n\nThrough an API you always get the exact string, because you have to type it:\n\n```bash\ncurl https://api.example.com/v1/chat/completions \\\n  -H \"Authorization: Bearer $KEY\" \\\n  -d '{\"model\": \"gpt-4o-mini-2024-07-18\", \"messages\": [...]}'\n```\n\n## Why the date on the end matters\n\nTwo names can point at the same family, the same version and the same size, and behave differently:\n\n- `claude-sonnet-4-5`: an alias. It resolves to that model's dated snapshot. Some providers' aliases move to a newer build without notice: OpenAI's `gpt-4o` has pointed at more than one dated snapshot.\n- `claude-sonnet-4-5-20250929`: a pinned snapshot. Same weights every day until it is retired (this one on 30 November 2026).\n\nAn alias is convenient and it is also how your carefully tuned workflow changes behaviour on a Tuesday morning without anybody deploying anything. If you are running something that matters, pin the date. If you are chatting, you cannot pin anything, so keep the possibility in mind when a tool seems to have changed personality.\n\nSnapshots do not live forever. Each provider publishes a retirement schedule. OpenAI promises at least six months' notice for generally available models. Anthropic lists a date before which each model will not be retired (about a year after its release) and then gives at least 60 days' notice, which in practice has been about two months. After retirement the name stops working and your script returns an error rather than a worse answer. That error is a kindness. The alias would have silently degraded instead.\n\n## The one thing a name never tells you\n\nA model name says nothing about the product wrapped around it. `gpt-4o` inside a chat app with web search, file upload and memory is a different experience from `gpt-4o` inside a badly built browser extension that sends three sentences of context. The name identifies the engine. It does not identify the car.\n\nSo the habit is: find the exact string, write it down with today's date next to your judgement of the tool, and remember that half of what you liked or disliked probably came from a layer the name does not cover."},"check":{"question":"A team pins their API calls to a dated snapshot rather than the family alias. Six months later the call starts returning an error saying the model no longer exists. What does this show about the choice they made?","options":[{"text":"The pin failed, because pinning was supposed to guarantee the model would remain available indefinitely.","correct":false,"tempting":"Confuses stability of behaviour with permanence of hosting; a snapshot promises the same weights while it is served, not that it is served forever."},{"text":"Dated snapshots are only for testing and production should always use the family alias so it stays current.","correct":false,"tempting":"Inverts the trade-off: the alias keeps working precisely by changing the weights underneath, which is the failure the pin was chosen to prevent."},{"text":"The pin worked: it traded a silent behaviour change for a loud failure on a published date.","correct":true},{"text":"The error means the provider withdrew the model early and the team should move to an open model they can host themselves.","correct":false,"tempting":"Jumps to a conclusion the evidence does not support; deprecation dates for pinned builds are published months ahead and an expiry is not an early withdrawal."}]},"prev":{"slug":"three-layers","title":"What you are actually choosing","url":"https://addaly.com/learn/choosing-your-tools/three-layers"},"next":{"slug":"sizes-and-tiers","title":"Small, medium, large: what the size label buys","url":"https://addaly.com/learn/choosing-your-tools/sizes-and-tiers"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Reading a model's name', in Choosing and Using the Tools. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/choosing-your-tools/reading-a-model-name (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Reading a model’s name","courseTitle":"Choosing and Using the Tools","module":{"index":1,"slug":"anatomy-of-a-tool","title":"What a tool is made of"},"answer":null,"keyIdea":"A model name encodes family, version, size class and build date, and the size class and the build date are the two parts that quietly decide what you actually get.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-name-is-a-specification","runs":[{"text":"The name is a specification"}]},{"t":"p","runs":[{"text":"claude-3-5-sonnet-20241022","code":true},{"text":". "},{"text":"gpt-4o-mini","code":true},{"text":". "},{"text":"gemini-2.5-flash","code":true},{"text":". "},{"text":"llama-3.1-70b-instruct","code":true},{"text":". "},{"text":"qwen3-8b","code":true},{"text":". "},{"text":"deepseek-r1","code":true},{"text":". These look like noise until you know the parts, and then each one tells you roughly what you are dealing with before you send a single message."}]},{"t":"p","runs":[{"text":"Almost every name carries four things, in some order."}]},{"t":"p","runs":[{"text":"The family.","b":true},{"text":" Who made it and which lineage it belongs to. GPT, Claude, Gemini, Llama, Mistral, Qwen, DeepSeek, Gemma, Phi."}]},{"t":"p","runs":[{"text":"The version.","b":true},{"text":" A number that goes up: 3, 3.5, 4, 4.1. A bigger number from the same maker is usually a better model, and this is the one part of the name you can mostly trust."}]},{"t":"p","runs":[{"text":"The size or speed label.","b":true},{"text":" This is the part people miss. "},{"text":"mini","code":true},{"text":", "},{"text":"nano","code":true},{"text":", "},{"text":"flash","code":true},{"text":", "},{"text":"lite","code":true},{"text":", "},{"text":"small","code":true},{"text":", "},{"text":"haiku","code":true},{"text":" mean the cheap fast one. "},{"text":"pro","code":true},{"text":", "},{"text":"opus","code":true},{"text":", "},{"text":"large","code":true},{"text":", "},{"text":"ultra","code":true},{"text":" mean the expensive slow one. "},{"text":"sonnet","code":true},{"text":", "},{"text":"medium","code":true},{"text":", and unlabelled names usually sit in the middle. Open models put the parameter count in directly: "},{"text":"8b","code":true},{"text":" is eight billion parameters, "},{"text":"70b","code":true},{"text":" is seventy."}]},{"t":"p","runs":[{"text":"The build or date.","b":true},{"text":" "},{"text":"20241022","code":true},{"text":" is a snapshot from 22 October 2024. "},{"text":"-preview","code":true},{"text":", "},{"text":"-exp","code":true},{"text":", "},{"text":"-latest","code":true},{"text":" are channels rather than dates, and they behave differently, which the last lesson of this module is about."}]},{"t":"p","runs":[{"text":"Two extra words appear on downloadable models. "},{"text":"instruct","code":true},{"text":" or "},{"text":"chat","code":true},{"text":" means it has been trained to follow instructions and hold a conversation: this is the one you want. A "},{"text":"base","b":true},{"text":" model with no such suffix continues text rather than answering it; ask it a question and it may write three more questions. People download the base model by mistake, conclude the model is broken, and give up."}]},{"t":"h2","id":"version-numbers-are-not-comparable-across-companies","runs":[{"text":"Version numbers are not comparable across companies"}]},{"t":"p","runs":[{"text":"Gemini 2.5 is not “worse than” GPT-4 because 2.5 is less than 4. The numbers are internal to each company and reset whenever marketing decides. OpenAI has shipped "},{"text":"gpt-4o","code":true},{"text":" and "},{"text":"o4-mini","code":true},{"text":" as different products in the same period; one is a general model, the other from a reasoning line. Read the family first, the number second, and never compare two makers’ version numbers directly."}]},{"t":"h2","id":"finding-out-what-is-actually-answering-you","runs":[{"text":"Finding out what is actually answering you"}]},{"t":"p","runs":[{"text":"In a chat product, look in three places, in this order:"}]},{"t":"ol","start":1,"items":[[{"text":"The model picker at the top of the conversation. It names a product tier, not always the exact model."}],[{"text":"Settings, then the account or model page."}],[{"text":"The small print under an individual answer: some products name the model there, especially after a fallback."}]]},{"t":"p","runs":[{"text":"If it says something like "},{"text":"”Fast”","i":true},{"text":", "},{"text":"”Smart”","i":true},{"text":", or "},{"text":"”Auto”","i":true},{"text":", you have been given a routing label rather than a model name. That is a real answer too: it means the product is choosing for you and may choose differently tomorrow."}]},{"t":"p","runs":[{"text":"Through an API you always get the exact string, because you have to type it:"}]},{"t":"code","lang":"bash","label":"Bash","text":"curl https://api.example.com/v1/chat/completions \\\n  -H \"Authorization: Bearer $KEY\" \\\n  -d '{\"model\": \"gpt-4o-mini-2024-07-18\", \"messages\": [...]}'"},{"t":"h2","id":"why-the-date-on-the-end-matters","runs":[{"text":"Why the date on the end matters"}]},{"t":"p","runs":[{"text":"Two names can point at the same family, the same version and the same size, and behave differently:"}]},{"t":"ul","items":[[{"text":"claude-sonnet-4-5","code":true},{"text":": an alias. It resolves to that model’s dated snapshot. Some providers’ aliases move to a newer build without notice: OpenAI’s "},{"text":"gpt-4o","code":true},{"text":" has pointed at more than one dated snapshot."}],[{"text":"claude-sonnet-4-5-20250929","code":true},{"text":": a pinned snapshot. Same weights every day until it is retired (this one on 30 November 2026)."}]]},{"t":"p","runs":[{"text":"An alias is convenient and it is also how your carefully tuned workflow changes behaviour on a Tuesday morning without anybody deploying anything. If you are running something that matters, pin the date. If you are chatting, you cannot pin anything, so keep the possibility in mind when a tool seems to have changed personality."}]},{"t":"p","runs":[{"text":"Snapshots do not live forever. Each provider publishes a retirement schedule. OpenAI promises at least six months’ notice for generally available models. Anthropic lists a date before which each model will not be retired (about a year after its release) and then gives at least 60 days’ notice, which in practice has been about two months. After retirement the name stops working and your script returns an error rather than a worse answer. That error is a kindness. The alias would have silently degraded instead."}]},{"t":"h2","id":"the-one-thing-a-name-never-tells-you","runs":[{"text":"The one thing a name never tells you"}]},{"t":"p","runs":[{"text":"A model name says nothing about the product wrapped around it. "},{"text":"gpt-4o","code":true},{"text":" inside a chat app with web search, file upload and memory is a different experience from "},{"text":"gpt-4o","code":true},{"text":" inside a badly built browser extension that sends three sentences of context. The name identifies the engine. It does not identify the car."}]},{"t":"p","runs":[{"text":"So the habit is: find the exact string, write it down with today’s date next to your judgement of the tool, and remember that half of what you liked or disliked probably came from a layer the name does not cover."}]}],"check":{"question":[{"text":"A team pins their API calls to a dated snapshot rather than the family alias. Six months later the call starts returning an error saying the model no longer exists. What does this show about the choice they made?"}],"options":[{"text":[{"text":"The pin failed, because pinning was supposed to guarantee the model would remain available indefinitely."}],"correct":false,"tempting":[{"text":"Confuses stability of behaviour with permanence of hosting; a snapshot promises the same weights while it is served, not that it is served forever."}]},{"text":[{"text":"Dated snapshots are only for testing and production should always use the family alias so it stays current."}],"correct":false,"tempting":[{"text":"Inverts the trade-off: the alias keeps working precisely by changing the weights underneath, which is the failure the pin was chosen to prevent."}]},{"text":[{"text":"The pin worked: it traded a silent behaviour change for a loud failure on a published date."}],"correct":true,"tempting":null},{"text":[{"text":"The error means the provider withdrew the model early and the team should move to an open model they can host themselves."}],"correct":false,"tempting":[{"text":"Jumps to a conclusion the evidence does not support; deprecation dates for pinned builds are published months ahead and an expiry is not an early withdrawal."}]}],"order":[3,0,2,1],"why":null},"prev":{"slug":"three-layers","title":"What you are actually choosing","minutes":7},"next":{"slug":"sizes-and-tiers","title":"Small, medium, large: what the size label buys","minutes":9}}}