11 ms·
Making 2.5 Flash and 2.5 Pro GA, and introducing Gemini 2.5 Flash-Lite
- nikanj 1y agoI need an AI model to be able to keep track of the AI model names.
- bigbuppo 1y ago[flagged]
- jjtheblunt 1y agoin a very real sense, Gemini (etc) _are_ fixed search.
- rootnod3 1y agoIf everyone searches through Gemini, then there are no click conversions. Without click conversions there is no incentive for many websites to make new content. Without content there is nothing new to learn for Gemini. It's an Ouroboros problem.
- ctoth 1y agoSo you mean only people who want to post something for the fun of posting it will post and we won't get as much corporate crap/SEO? Darn. Of course practically this means people who post will have a reason to post and that reason might be to influence Gemini, so double darn.
- rootnod3 1y agoI said a majority. Say stack overflow, medium articles, online newspapers, they live off of click conversion. The internet is more than people just posting for fun. Of course everyone only posting for fun would be ideal. But even if you only post for fun, if most people only find your content by seeing it re-hashed by a LLM that hallucinates some other stuff in the middle, I'd probably also lose the fun in posting for posting sake.
- HotHotLava 1y agoFor me, I actually feel much more motivated to write good and accurate documentation knowing there will be at least one reader who is going to look at it very closely and will attempt to synthesize useful information from it. Same with my old open-source projects, it's kinda cool knowing that all the old stuff that nobody would have ever looked at anymore is now part of a humanity-wide corpus of useful knowledge on how to do X with language Y.
- jsnider3 1y agoYeah, a lot of the web is at-risk if searchers just read LLM summaries and don't click through, but we will have Skynet before that becomes a real issue and then this will all be irrelevant.
- bigbuppo 1y agoFortunately it was trained on reddit comments. Oh no.
- wredcoll 1y agoStack overflow makes a profit off click conversions. It does not exist because of it. People would have built a version regardless of the profit margin involved.
- rootnod3 1y agoRunning the site costs money. So you either run ads or make users pay. Either way, usage of LLMs for searching would decrease the income for these sites.
- jjtheblunt 1y agoKagi is exactly a counterexample to your assertion.
- recursive 1y agoYou seem to be assuming that there will be no paid ads in Gemini output. It's hard for me to believe that is going to be true in five years.
- rootnod3 1y agoOh, there's definitely going to be ads in there. Probably not even marked or visible as such. I fear that it's going to be weaved into the text somewhere.
- krembo 1y agoNobody cares about search anymore
- remus 1y agoI mean the model names are always a bit odd, but flash-lite is particulary good!
- wayeq 1y agokinda sounds like something else
- candiddevmike 1y agoCurious to hear what folks are doing with Gemini outside of the coding space and why you chose it. Are you building your app so you can swap the underlying GenAI easily? Do you "load balance" your usage across other providers for redundancy or cost savings? What would happen if there was ever some kind of spot market for LLMs?
- crowcroft 1y agoSimple unstructured to structured data transformation. I find Flash and Flash Lite are more consistent than others as well as being really fast and cheap. I could swap to other providers fairly easily, but don't intend to at this point. I don't operate at a large scale.
- willidiots 1y agoLow-latency LLM for my home automation. Anecdotally, Gemini was much quicker than OpenAI in responding to simple commands. In general, when I need "cheap and fast" I choose Gemini.
- thimabi 1y agoIn my experience, Gemini 2.5 Pro really shines in some non-coding use cases such as translation and summarization via Canvas. The gigantic context window and large usage limits help in this regard. I also believe Gemini is much better than ChatGPT in generating deep research reports. Google has an edge in web search and it shows. Gemini’s reports draw on a vast number of sources, thus tend to be more accurate. In general, I even prefer its writing style, and I like the possibility of exporting reports to Google Docs. One thing that I don’t like about Gemini is its UI, which is miles behind the competition. Custom instructions, projects, temporary chats… these things either have no equivalent in Gemini or are underdeveloped.
- hnuser123456 1y agoIf you're a power user, you should probably be using Gemini through AI studio rather than the "basic user" version. That allows you to set system instructions, temperature, structured output, etc. There's also NotebookLM. Google seems to be trying to make a bunch of side projects based on Gemini and seeing what sticks, and the generic gemini app/webchat is just one of those.
- varun_chopra 1y agoAt one point, when they made Gemini Pro free on AI Studio, Gemini was the model of choice for many people, I believe. Somehow it's gotten worse since then, and I'm back to using Claude for serious work. Gemini is like that guy who keeps talking but has no idea what he's actually talking about. I still use Gemini for brainstorming, though I take its suggestions with several grains of salt. It's also useful for generating prompts that I can then refine and use with Claude.
- chrismustcode 1y agoWhen I ask it do to do something in cursor it goes full sherlock thinking about every possible outcome. Just claude 4 sonnet with thinking just has a bit think then does it
- UncleOxidant 1y agoUsed to be able to use Gemini Pro free in cline. Now the API limits are so low that you immediately get messages about needing to top up your wallet and API queries just don't go through. Back to using DeepSeek R1 free in cline (though even that eventually stops after a few hours and you have to wait until the next day for it to work again). Starting to look like I need to setup a local LLM for coding - which means it's time to seriously upgrade my PC (well, it's been about 10 years so it was getting to be time anyway)
- Workaccount2 1y agoBy the time you breakeven on whatever you spend on a decent LLM capable build, your hardware will be too far behind to run whatever is best locally then. It's something that feels cheaper but with the pace of things, unless you are churning an insane amount of tokens, probably doesn't make sense. Never mind that local models running on 24 or 48GB are maybe around flash-lite in ability while being slower than SOTA models. Local models are mostly for hobby and privacy, not really efficiency.
- FirmwareBurner 1y agoI found Gemini now terrible for coding. I gave it my code blocks and told it what to change and it added tonnes and tonnes of needles extra code plus endless comments. It turned a tight code into a Papyrus. ChatGPT is better but tends to be too agreeable, never trying to disagree with what you say even if it's stupid so you end up shooting yourself in the foot. Claude seems like the best compromise. Just my two kopecks.
- jbellis 1y agoLove to see it, this takes Flash Lite from "don't bother" territory for writing code to potentially useful. (Besides being inexpensive, Flash Lite is fast -- almost always sub-second, to as low as 200ms. Median around 400ms IME.) Brokk (https://brokk.ai/ https://brokk.ai/) currently uses Flash 2.0 (non-Lite) for Quick Edits, we'll evaluate 2.5 Lite now. ETA: I don't have a use case for a thinking model that is dumber than Flash 2.5, since thinking negates the big speed advantage of small models. Curious what other people use that for.
- MaxLeiter 1y agoTo me, if it thinks fast enough, I don't care how much thinking it does.
- hs86 1y agoI am always disappointed when I compare the answers to the same queries on 2.5 Pro vs. o4-mini/o3. But trying out the same query in AI Studio gives much better results, closer to OpenAI's models. What is wrong with 2.5 Pro in the Gemini app? I can't believe that the model in their consumer app would produce the same benchmark results as 2.5 Pro in the API or AI Studio.
- mh- 1y agoI don't have any inside information, but I'm sure there are different system prompts used in the Gemini chat interface vs the API. On OpenAI/ChatGPT they're sometimes dramatically different.
- thimabi 1y agoThe models in the Gemini app are nerfed in comparison to those in AI Studio: they have less thinking budget, output less tokens, and have various safety filters. There’s certainly a trade-off between using AI Studio for its better performance and using the API or the Gemini app in a way that doesn’t involve Google keeping your data for training purposes.
- 2Gkashmiri 1y agoI have a huge background.js file from a now removed browser extension that the Devs made into a single line. Around 800KB of a single line file I think.... I tried many free stuff to try to refactor it but they all loose context window quickly.
- mh- 1y agoThere are myriad non-LM tools that can deobfuscate and prettify JS. I used them with success long before LLMs were en vogue.
- 2Gkashmiri 1y agoLink?
- mh- 1y agoCheck out http://www.jsnice.org/ http://www.jsnice.org/ and google javascript deobfuscator for others.
- BonoboIO 1y agoWhich extension is it?
- 2Gkashmiri 1y agoGstzen : peaceful compliance. Not in store right now,
- sergiotapia 1y agoConsidering moving from Groq Llama 3.3 70b to Gemini 2.5 Flash Lite for one of my use cases. Results are coming in great, and it's very fast (important for my real-time user perception needs). What kind of rate limits do these new Gemini models have?
- UncleOxidant 1y agoAre you using Groq Llama 3.3 70b from something like cline? Is it free and what are the API query limits?
- sergiotapia 1y agoI'm using it from their HTTP API. Limits I can't remember what they were initially tbh, I had to reach out through backchannels to get it increased to 300,000 tokens per minute.
- serjester 1y agoI'm glad that they standardized pricing for the thinking vs non-thinking variant. A couple weeks ago I accidentally spent thousands of extra dollars by forgetting to set the thinking budget to zero. Forgetting a single config parameter should not automatically raise the model cost 5X. [edit] I'm less excited about this because it looks like their solution was to dramatically raise the base price on the non-thinking variant.
- heliophobicdude 1y agoWishing they release the Gemini Diffusion model. It'll quickly replace the default model for Aider.
- causal 1y agoWhy do you think so? I've played with the Diffusion model a bit and it makes a lot of mistakes
- heliophobicdude 1y agoIt's super fast to iterate with? Curious what mistakes are you seeing and how were you using it? I use the instant edit feature with directions on exact code changes.
- vessenes 1y agoIt feels to me like properly instrumented, these diffusion models are going to be really powerful coding tools. Imagine a “smart” model carving out a certain number of tokens in a response for each category of response output, then diffusing the categories.
- GaggiX 1y ago2.5 Flash Lite seems better at everything compare to 2.0 Flash Lite with the only exception being SimpleQA, so there is probably a small tradeoff on pop culture knowledge for coding, math, science, reasoning and multimodal tasks.
- simonw 1y agoThey don't mention it in the post, but it looks like this includes a price increase for the Gemini 2.5 Flash model. For 2.5 Flash Preview https://web.archive.org/web/20250616024644/https://ai.google.dev/gemini-api/docs/pricing https://web.archive.org/web/20250616024644/https://ai.google... $0.15/million input text / image / video $1.00/million audio Output: $0.60/million non-thinking, $3.50/million thinking The new prices for Gemini 2.5 Flash ditch the difference between thinking and non-thinking and are now: https://ai.google.dev/gemini-api/docs/pricing https://ai.google.dev/gemini-api/docs/pricing $0.30/million input text / image / video (2x more) $1.00/million audio (same) $2.50/million output - significantly more than the old non-thinking price, less than the old thinking price.
- rudedogg 1y agoA cool 2x+ price increase. And Gemini 2.0 Flash was $0.10/$0.40.
- k8sToGo 1y agoYou can also see this difference in open router. But why is there only thinking flash now?
- hnuser123456 1y agoApparently, you can make a request to 2.5 flash to not use thinking, but it will still sometimes do it anyways, this has been an issue for months, and hasn't been fixed by model updates: https://github.com/google-gemini/cookbook/issues/722 https://github.com/google-gemini/cookbook/issues/722
- zelias 1y agoNot sure where else to post this, but when attempting to use any of the Gemini 2.5 models via API, I receive an "empty content" response about 50% of the time. To be clear, the API responds successfully, but the `content` returned by the LLM is just an empty string. Has anyone here had any luck working around this problem?
- Tiberium 1y agoWhat finish reason are you getting? Perhaps your code sets a low max_tokens, so the generation stops while the model is still thinking, without giving any actual output.
- zelias 1y agoThe finish reason is `length`. I have tried setting minimal token budgets, really small prompts, and max lengths of various sizes from 100-4000 and nothing seems to make a consistent dent in the behavioral pattern.
- danbrooks 1y agoThis can happen if the prompt or response is blocked by a safety filter. Check some of the other fields in the response.
- zzleeper 1y agoGood luck using 2.5 for anything non-trivial. I have about 500,000 news articles I am parsing. OpenAI models work well but found Gemini had fewer mistakes. Problem is; they give me a terrible 10k RPD limit. To increase to the next tier, they then require a minimum amount of spending but I can't reach that amount even when maxing the RPD limit for multiple days in a row. I emailed them twice and completed their forms but everyone knows how this works. So now I'm back at OpenAI, with a model with a bit more mistakes but that won't 403 me after half an hour of using it due to their limits.
- eldenring 1y agoI'm guessing now that it is GA this won't be a problem.
- zzleeper 1y agoI wish! The tier-based limits are still the same! At least it's more expensive now so I guess I will be able to hop to the next tier sooner? ¯\_(ツ)_/¯
- be7a 1y agoThe rate limits apply only to the Gemini API. There is also Vertex from GCP, which offers the same models (and even more, such as Claude) at the same pricing, but with much higher rate limits (basically none, as long as they don't need to cut anyone off with provisioned throughput iiuc) and with a process to get guaranteed throughput.
- zzleeper 1y agoHad no idea... always thought Vertex was just a way to do enterprise offering!
- b0a04gl 1y ago[dead]
- lvl155 1y agoI am very impressed with Gemini and stopped using OpenAI. Sometimes, I ping all three major models on OpenRouter but 90% is on Gemini now. Compare that to 90% ChatGPT last year.
- aatd86 1y agoSame. For now I have canceled my claude subscription. Gemini has been catching up.
- glohbalrob 1y agoAlso me. Still pay for OpenAI, I use gpt4 for excel work and is super fast and able to do more excel related work like combine files that come up often for projects I work on.
- codingwagie 1y agoI love to hate on google, but yeah their models are really good. The larger context window is huge
- kapildev 1y agoDoesn't OpenAI's GPT 4.1 also have 1 million context length?
- voiper1 1y agoI don't like the thinking time, but for coding, journaling, and other stuff I've often been impressed with Gemini Pro 2.5 out of the box. Possibly I could do much more prompt fine-tuning to nudge openai/anthropic in the direction I want, but with the same prompts Gemini often gives me answers/structure/tone I like much better. Example: I had claude 3.7 generating embedding images and captions along with responses. Same prompt into Gemini it gave much more varied and flavorful pictures.
- deanstag 1y agoCurious. How do you use gemini for journaling? What is your workflow?
- tiahura 1y agoGemini 2.5 doesn’t get enough credit for the quality of its writing in non-code (eg law) topics. It’s definitely a notch below Claude 4, but well ahead of ChatGPT 4o, 4.5, o3.
- jjani 1y agoClassic bait-and-switch to make developers build things on top off models for 2 months, and then raise input price by 2x and output by 4x. But hey, it's Google, wouldn't expect anything else from an advertising company.
- zurfer 1y agofor anyone, who was expecting more news: the GA models benchmark basically the same as the last preview models. It's really just Google telling us that we get less api errors and this model will have a checkpoint for a longer time.
- dinesh2609 1y ago6.33X increase in the price of Audio processing compared to 2.0 Flash-Lite Gemini 2.5 Flash Lite (Audio Input) - $0.5/million tokens Gemini 2.0 Flash Lite (Audio Input) - $0.075/million tokens Wonder what led to such a high bump in Audio token processing
- sethkim 1y agoI run a batch inference/LLM data processing service and we do a lot of work around cost and performance profiling of (open-weight) models. One odd disconnect that still exists in LLM pricing is the fact that providers charge linearly with respect to token consumption, but costs are actually quadratic with an increase in sequence length. At this point, since a lot of models have converged around the same model architecture, inference algorithms, and hardware - the chosen costs are likely due to a historical, statistical analysis of the shape of customer requests. In other words, I'm not surprised to see costs increase as providers gather more data about real-world user consumption patterns.
- diziet 1y agoAren't advances in KV caching making compute cost not quite quadratic?
- raybb 1y agoIt's a bummer that 2.5 Pro is still removed from the free tier of the API.
- zhyder 1y agoBlended price (assuming 3:1 for input:output tokens) is 3.24x of what was stated before [1], and now nearly 5x of 2.0 Flash. Makes 2.0 Flash a still competitive option for many use-cases, particularly ones that aren't coding-heavy I think. A slightly poorer performing model can net perform better through multiple prompt passes. Bummer, was hoping 2.5 Flash would be a slam dunk choice. [1] - https://web.archive.org/web/20250616024644/https://ai.google.dev/gemini-api/docs/pricing https://web.archive.org/web/20250616024644/https://ai.google...
- andrewstuart 1y agoGemini strangely says you cannot upload all sorts of file types. But it accepts them just fine if you upload a zip file……. which you can only do in AI studio.
- andrewstuart 1y agoI really wish all the AI companies would down tools on all development until they work out file downloads, ftp, sftp, git ANY way to access the files other than copy paste and “download file”. The workflow is crushingly tedious. And no I don’t want to use an AI IDE or some other tool. I like the UI of Gemini chat and AI Studio and I want them improved.
- simonw 1y agoI tried using the three new models to transcribe the audio of this morning's Gemini Twitter Space. I got very strong results from 2.5 Pro and 2.5 Flash, but 2.5 Flash Lite sadly got stuck in a loop until it ran out of output tokens: Um, like, what did the cows bring to you? Nothing. And then, um, and then, uh, and then, uh, and then, uh, and then, uh, and then, uh, and then, uh, and then, uh, and... Notes on my results (including the transcripts which worked, which included timestamps and guessed speaker names) here: https://simonwillison.net/2025/Jun/17/gemini-2-5/#transcribing-audio-from-a-twitter-space https://simonwillison.net/2025/Jun/17/gemini-2-5/#transcribi...
- synergy20 1y agoi cancelled chatgpt early this year and switched to Gemini, with Gemini making progress rapidly, i wonder if openai already lost the battle
- dhao 1y agoAnyone else unable to access 2.5-pro via api? I'm currently getting "Publisher Model `projects/349775993245/locations/us-west4/publishers/google/models/gemini-2.5-pro` was not found or your project does not have access to it. Please ensure you are using a valid model version."
- davhao 1y agoAnyone else unable to access 2.5-pro via api? I'm currently getting "Publisher Model `projects/349775993245/locations/us-west4/publishers/google/models/gemini-2.5-pro` was not found or your project does not have access to it. Please ensure you are using a valid model version."
- atlgator 1y agoIs there a Codex/Claude Code competitor on the way?
- Ninjinka 1y agoJules?
- ziofill 1y agoI dream of a day when LLM naming follows a convention.
- DidYaWipe 1y agoWhich are what?
- jl6 1y agoI had a great-ish result from 2.5 Pro the other day. I asked it to date an old photograph, and it successfully read the partial headline on a newspaper in the background (which I had initially thought was too small/blurry to make out) and identified the 1980s event it was reporting. Impressive. But then it confidently hallucinated the date of the article (which I later verified by checking in an archive).
- Tixx7 1y agoI switched to 2.5 Flash (non-think) for most of my projects because it was such a good model with good pricing. Cost is an important factor so hoping that flash-lite is sufficient, even tho its somtimes more than 50% worse in relevant benchmarks which sucks. Was also just looking at 4.1-mini but thats more expensive and often scores around the same as flash-lite in benchmarks (except coding which i dont care about). Crazy to think that even after this move by google, openai is still the worse option for me, at least regarding API. Other than API im actually using chatgpt (o3/o4-mini, 4o is a joke) a lot more again lately after 2.5 Pro got nerfed