github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-15 12:48:05 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #11766] Support reasoning effort for gpt-oss #85482

New Issue
Closed
opened 2026-05-10 00:17:35 -05:00 by GiteaMirror · 25 comments
GiteaMirror commented 2026-05-10 00:17:35 -05:00
Owner
Copy Link

Originally created by @ericssonl07 on GitHub (Aug 7, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11766

Originally assigned to: @hoyyeva on GitHub.

Currently unable to adjust reasoning effort between "low" "medium" and "high"- would be very helpful as a special feature for gpt-oss (if any other ollama models support, that would be nice too)

Originally created by @ericssonl07 on GitHub (Aug 7, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11766 Originally assigned to: @hoyyeva on GitHub. Currently unable to adjust reasoning effort between "low" "medium" and "high"- would be very helpful as a special feature for gpt-oss (if any other ollama models support, that would be nice too)
GiteaMirror added the appfeature request labels 2026-05-10 00:17:36 -05:00
GiteaMirror commented 2026-05-10 00:17:38 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 7, 2025):

#11752

<!-- gh-comment-id:3162395290 --> @rick-github commented on GitHub (Aug 7, 2025): #11752
GiteaMirror commented 2026-05-10 00:17:38 -05:00
Author
Owner
Copy Link

@ericssonl07 commented on GitHub (Aug 7, 2025):

@rick-github in the app, not in the API endpoint

<!-- gh-comment-id:3162466989 --> @ericssonl07 commented on GitHub (Aug 7, 2025): @rick-github in the app, not in the API endpoint
GiteaMirror commented 2026-05-10 00:17:39 -05:00
Author
Owner
Copy Link

@Mondonno commented on GitHub (Aug 28, 2025):

Seconding this.

<!-- gh-comment-id:3233604064 --> @Mondonno commented on GitHub (Aug 28, 2025): Seconding this.
GiteaMirror commented 2026-05-10 00:17:39 -05:00
Author
Owner
Copy Link

@moontato commented on GitHub (Nov 21, 2025):

thirding this

<!-- gh-comment-id:3561362204 --> @moontato commented on GitHub (Nov 21, 2025): thirding this
GiteaMirror commented 2026-05-10 00:17:40 -05:00
Author
Owner
Copy Link

@hoyyeva commented on GitHub (Dec 10, 2025):

Closing this issue as it is already being addressed. Now you are able to select different level of reasoning for all the gpt-oss models

<!-- gh-comment-id:3639275736 --> @hoyyeva commented on GitHub (Dec 10, 2025): Closing this issue as it is already being addressed. Now you are able to select different level of reasoning for all the gpt-oss models
GiteaMirror commented 2026-05-10 00:17:41 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Feb 7, 2026):

Closing this issue as it is already being addressed. Now you are able to select different level of reasoning for all the gpt-oss models

Still getting level=WARN source=types.go:966 msg="invalid option provided" option=reasoning_effort floods with 0.15.5.

<!-- gh-comment-id:3864082412 --> @OdinVex commented on GitHub (Feb 7, 2026): > Closing this issue as it is already being addressed. Now you are able to select different level of reasoning for all the gpt-oss models Still getting `level=WARN source=types.go:966 msg="invalid option provided" option=reasoning_effort` floods with `0.15.5`.
GiteaMirror commented 2026-05-10 00:17:41 -05:00
Author
Owner
Copy Link

@chigkim commented on GitHub (Mar 1, 2026):

Is it possible to set reasoning effort with openai api endpoint?
If so, is it possible to set reasoning effort to "none", and entirely disable thinking for other models like qwen3.5 as well?

<!-- gh-comment-id:3978899530 --> @chigkim commented on GitHub (Mar 1, 2026): Is it possible to set reasoning effort with openai api endpoint? If so, is it possible to set reasoning effort to "none", and entirely disable thinking for other models like qwen3.5 as well?
GiteaMirror commented 2026-05-10 00:17:42 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 1, 2026):

@chigkim reasoning_effort="none"

<!-- gh-comment-id:3981230931 --> @rick-github commented on GitHub (Mar 1, 2026): @chigkim `reasoning_effort="none"`
GiteaMirror commented 2026-05-10 00:17:42 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

Still getting this with 0.18.2. source=types.go:976 msg="invalid option provided" option=reasoning_effort, value doesn't make a difference.

<!-- gh-comment-id:4107096668 --> @OdinVex commented on GitHub (Mar 22, 2026): Still getting this with `0.18.2`. `source=types.go:976 msg="invalid option provided" option=reasoning_effort`, value doesn't make a difference.
GiteaMirror commented 2026-05-10 00:17:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

reasoning_effort for OpenAI API. think for Ollama API.

<!-- gh-comment-id:4107099306 --> @rick-github commented on GitHub (Mar 22, 2026): `reasoning_effort` for OpenAI API. `think` for Ollama API.
GiteaMirror commented 2026-05-10 00:17:44 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

reasoning_effort for OpenAI API. think for Ollama API.

Think and Reasoning Effort are not the same.
Reasoning Effort: https://developers.openai.com/api/docs/guides/reasoning
Think: https://docs.ollama.com/capabilities/thinking

Reasoning Effort literally affects how many tokens will be allocated, a guidance on actual "effort". Think just shows thought process (such as "The user is requesting this, we should say Blank. Response: Blank").

<!-- gh-comment-id:4107105456 --> @OdinVex commented on GitHub (Mar 22, 2026): > `reasoning_effort` for OpenAI API. `think` for Ollama API. Think and Reasoning Effort are not the same. Reasoning Effort: https://developers.openai.com/api/docs/guides/reasoning Think: https://docs.ollama.com/capabilities/thinking Reasoning Effort literally affects how many tokens will be allocated, a guidance on actual "effort". Think just shows thought process (such as "The user is requesting this, we should say Blank. Response: Blank").
GiteaMirror commented 2026-05-10 00:17:44 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

think:false would be same as reasoning_effort:none if gpt-oss supported turning off thinking. think:true is the same as reasoning_effort:medium.

$ for t in low medium high ; do r="$(curl -s localhost:11434/v1/chat/completions -d '{"model":"gpt-oss:20b","messages":[{"role":"user","content":"why is the sky blue"}],"reasoning_effort":"'$t'","stream":false}' | jq .choices[0].message.reasoning)" ; printf "%-7s %4d %s\n" $t $(wc -c <<< "$r") "$r" ; done
low       33 "Need explain color scattering."
medium   349 "The user asks: \"why is the sky blue\". Classic explanation: Rayleigh scattering. Provide some details. Provide also mention about wavelength, scattering intensity, sky color variations at sunrise/sunset, etc. Should be clear and concise. Provide some analogies maybe. Answer in plain language. Should be good for general audience. Let's produce."
high    5972 "We need to answer the question: \"why is the sky blue\". The user asks presumably a general knowledge question. I need to respond comprehensively.\n\nWe should cover Rayleigh scattering explanation: shorter wavelengths (blue/violet) scatter more, earth's atmosphere scatters Blue more; at sunrise/sunset, green and red due to longer path lengths. Explain that violet is scattered even more but human eye less sensitive, and violet is absorbed by ozone. Also mention that sky appears white at some times, the variation due to cloud presence, etc.\n\nProvide concise but thorough explanation: physics of scattering, types of scattering, maybe mention Mie scattering for dust, Rayleigh for molecules. Also mention scattering cross-section ∝ λ^-4. Explain why sky is blue: due to sunlight, the spectrum, scattering path. Show that the intensity of blue is higher. Also mention the phenomenon of Tyndall effect. Could mention that the sky appears blue for most of day because of scattering.\n\nAlso mention that the sky is the opposite of the ground. The sky appears blue because light leaves the Sun, scatters, the photons are scattered off atmospheric molecules, the scattered white light appears blue. The \"sky\" isn't actually scattering itself but the view of the atmosphere. The sky may appear violet at times but not because we are less sensitive. Show that our vision is more sensitive to blue-green. The sky appears white in cloudy days due to multiple scattering.\n\nExplain historically that the explanation of blue sky came about in 19th century with James Clerk Maxwell. Maybe show experiments.\n\nWe can mention that the reason is also due to the Rayleigh scattering phenomenon, discovered by Lord Rayleigh in 1871. Could provide references or link to known sources.\n\nAlso explain: The sun is a nearly perfect blackbody at 5,500K, its spectral distribution is roughly white with peaks. The atmosphere acts to filter the sunlight: Short wavelengths are heavily scattered; long wavelengths pass. The scattered blue light is what we see when looking at the sky. The light reflected is due to the air molecules and some fine dust.\n\nAlso mention that the scattering intensity is proportional to 1/(λ^4). The reason we don't see violet might be due to the absorption by ozone (UV), the eye's sensitivity.\n\nExplain also that the sky is tinted because of the composition: oxygen, nitrogen, small particles. Provide a conceptual explanation.\n\nWe could mention that at sunrise or sunset, the scattering path is longer, so the shorter wavelengths are scattered out leaving reds. So sunrise/sunset get red/orange hues.\n\nAlso mention that blue sky can be considered a \"Rayleigh-scattered\" phenomenon.\n\nAlso mention that if we remove the atmosphere or look at an extraterrestrial viewpoint from space, we can't see a blue sky. The sky would be black.\n\nAlso mention that there are also other factors: water vapor, microparticles scatter also light. The scattering from larger particles is Mie scattering which does not strongly depend on λ. So cloud scattering results in white.\n\nAlso we might discuss that the reason we perceive the sky as blue is due to the distribution of wavelengths we see.\n\nWe can also mention that the Earth has nitrogen (~78%), oxygen (~21%), Argon (~0.93%), CO2 (~0.04%) etc. The Rayleigh scattering cross-section for each gas is dependent on the refractive index. That can be explained.\n\nAlso mention how the scattering by the air molecules is anisotropic, forward peaked and back scattering may produce a bright sky near the sun.\n\nOk, but the user only asked: \"why is the sky blue\". It would be best to deliver a clear explanation. Possibly some details about how the sun's ray travels, interacts with atmosphere, scattering. Provide analogies: Like when you see a bright white light through clouds.\n\nWe should also mention that we see blue because when light passes through atmospheric molecules, Rayleigh scattering is stronger at shorter wavelengths (blue/violet), so the light leaving the atmosphere in the direction of the observer is richer in those wavelengths.\n\nAlso mention that the reason violet is rarely visible: Ozone absorbs more at violet wavelengths; our eyes are less sensitive in that region; plus after scattering, the color that remains is blue.\n\nAlso mention that if you removed the atmosphere, you would see a black sky, not blue.\n\nThe answer should also be accessible but not overly technical.\n\nWill produce a friendly answer with some bullet points.\n\nI also need to consider any safety or policy issues. This is a factual question. No disallowed content. No privacy concerns. The user didn't ask for any disallowed content.\n\nThus, proceed with answer.\n\nWe can optionally provide an example of the equation:\n\nI(λ) ∝ 1/λ^4.\n\nWe can also mention the effect of atmospheric aerosols causing white sky.\n\nAlso mention that in the early morning, if there is high humidity or dust, the scattering might produce a milder blue, etc.\n\nAlso mention that the reason the sky isn't bright blue at midday is due to scattering, that the scattering is less intense at midday? Actually midday the sun is overhead, so the scattering path is minimal. But the sky still is blue. The scattering is less intense at midday because less path length? The difference is relatively small; the sky brightens overall.\n\nAlso mention that Rayleigh scattering is dependent on molecular size relative to wavelength.\n\nOk. Let's answer.\n\nAdditionally, we might mention that scattering is the same phenomenon that makes the ocean look blue: water molecules scatter blue light.\n\nOk, I'll produce answer with explanation: Rayleigh scattering; human perception; and maybe some historical context. Provide also potential explanation about other reasons, such as sunlight spectral distribution, but mainly scattering.\n\nAnswer in friendly tone.\n\nShould I show image? Not necessary.\n\nOk writing."
<!-- gh-comment-id:4107137656 --> @rick-github commented on GitHub (Mar 22, 2026): `think:false` would be same as `reasoning_effort:none` if gpt-oss supported turning off thinking. `think:true` is the same as `reasoning_effort:medium`. ```console $ for t in low medium high ; do r="$(curl -s localhost:11434/v1/chat/completions -d '{"model":"gpt-oss:20b","messages":[{"role":"user","content":"why is the sky blue"}],"reasoning_effort":"'$t'","stream":false}' | jq .choices[0].message.reasoning)" ; printf "%-7s %4d %s\n" $t $(wc -c <<< "$r") "$r" ; done low 33 "Need explain color scattering." medium 349 "The user asks: \"why is the sky blue\". Classic explanation: Rayleigh scattering. Provide some details. Provide also mention about wavelength, scattering intensity, sky color variations at sunrise/sunset, etc. Should be clear and concise. Provide some analogies maybe. Answer in plain language. Should be good for general audience. Let's produce." high 5972 "We need to answer the question: \"why is the sky blue\". The user asks presumably a general knowledge question. I need to respond comprehensively.\n\nWe should cover Rayleigh scattering explanation: shorter wavelengths (blue/violet) scatter more, earth's atmosphere scatters Blue more; at sunrise/sunset, green and red due to longer path lengths. Explain that violet is scattered even more but human eye less sensitive, and violet is absorbed by ozone. Also mention that sky appears white at some times, the variation due to cloud presence, etc.\n\nProvide concise but thorough explanation: physics of scattering, types of scattering, maybe mention Mie scattering for dust, Rayleigh for molecules. Also mention scattering cross-section ∝ λ^-4. Explain why sky is blue: due to sunlight, the spectrum, scattering path. Show that the intensity of blue is higher. Also mention the phenomenon of Tyndall effect. Could mention that the sky appears blue for most of day because of scattering.\n\nAlso mention that the sky is the opposite of the ground. The sky appears blue because light leaves the Sun, scatters, the photons are scattered off atmospheric molecules, the scattered white light appears blue. The \"sky\" isn't actually scattering itself but the view of the atmosphere. The sky may appear violet at times but not because we are less sensitive. Show that our vision is more sensitive to blue-green. The sky appears white in cloudy days due to multiple scattering.\n\nExplain historically that the explanation of blue sky came about in 19th century with James Clerk Maxwell. Maybe show experiments.\n\nWe can mention that the reason is also due to the Rayleigh scattering phenomenon, discovered by Lord Rayleigh in 1871. Could provide references or link to known sources.\n\nAlso explain: The sun is a nearly perfect blackbody at 5,500K, its spectral distribution is roughly white with peaks. The atmosphere acts to filter the sunlight: Short wavelengths are heavily scattered; long wavelengths pass. The scattered blue light is what we see when looking at the sky. The light reflected is due to the air molecules and some fine dust.\n\nAlso mention that the scattering intensity is proportional to 1/(λ^4). The reason we don't see violet might be due to the absorption by ozone (UV), the eye's sensitivity.\n\nExplain also that the sky is tinted because of the composition: oxygen, nitrogen, small particles. Provide a conceptual explanation.\n\nWe could mention that at sunrise or sunset, the scattering path is longer, so the shorter wavelengths are scattered out leaving reds. So sunrise/sunset get red/orange hues.\n\nAlso mention that blue sky can be considered a \"Rayleigh-scattered\" phenomenon.\n\nAlso mention that if we remove the atmosphere or look at an extraterrestrial viewpoint from space, we can't see a blue sky. The sky would be black.\n\nAlso mention that there are also other factors: water vapor, microparticles scatter also light. The scattering from larger particles is Mie scattering which does not strongly depend on λ. So cloud scattering results in white.\n\nAlso we might discuss that the reason we perceive the sky as blue is due to the distribution of wavelengths we see.\n\nWe can also mention that the Earth has nitrogen (~78%), oxygen (~21%), Argon (~0.93%), CO2 (~0.04%) etc. The Rayleigh scattering cross-section for each gas is dependent on the refractive index. That can be explained.\n\nAlso mention how the scattering by the air molecules is anisotropic, forward peaked and back scattering may produce a bright sky near the sun.\n\nOk, but the user only asked: \"why is the sky blue\". It would be best to deliver a clear explanation. Possibly some details about how the sun's ray travels, interacts with atmosphere, scattering. Provide analogies: Like when you see a bright white light through clouds.\n\nWe should also mention that we see blue because when light passes through atmospheric molecules, Rayleigh scattering is stronger at shorter wavelengths (blue/violet), so the light leaving the atmosphere in the direction of the observer is richer in those wavelengths.\n\nAlso mention that the reason violet is rarely visible: Ozone absorbs more at violet wavelengths; our eyes are less sensitive in that region; plus after scattering, the color that remains is blue.\n\nAlso mention that if you removed the atmosphere, you would see a black sky, not blue.\n\nThe answer should also be accessible but not overly technical.\n\nWill produce a friendly answer with some bullet points.\n\nI also need to consider any safety or policy issues. This is a factual question. No disallowed content. No privacy concerns. The user didn't ask for any disallowed content.\n\nThus, proceed with answer.\n\nWe can optionally provide an example of the equation:\n\nI(λ) ∝ 1/λ^4.\n\nWe can also mention the effect of atmospheric aerosols causing white sky.\n\nAlso mention that in the early morning, if there is high humidity or dust, the scattering might produce a milder blue, etc.\n\nAlso mention that the reason the sky isn't bright blue at midday is due to scattering, that the scattering is less intense at midday? Actually midday the sun is overhead, so the scattering path is minimal. But the sky still is blue. The scattering is less intense at midday because less path length? The difference is relatively small; the sky brightens overall.\n\nAlso mention that Rayleigh scattering is dependent on molecular size relative to wavelength.\n\nOk. Let's answer.\n\nAdditionally, we might mention that scattering is the same phenomenon that makes the ocean look blue: water molecules scatter blue light.\n\nOk, I'll produce answer with explanation: Rayleigh scattering; human perception; and maybe some historical context. Provide also potential explanation about other reasons, such as sunlight spectral distribution, but mainly scattering.\n\nAnswer in friendly tone.\n\nShould I show image? Not necessary.\n\nOk writing." ```
GiteaMirror commented 2026-05-10 00:17:45 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

for t in low medium high ... "reasoning_effort":"'$t'" <-- Looks like you're using reasoning_effort, not think.

The output being different (much better with high) is exactly what I'm talking about, too. think only controls what is seen for trace, not whether the response has higher reasoning.

Think and Reasoning Effort are not the same.
Reasoning Effort: https://developers.openai.com/api/docs/guides/reasoning
Think: https://docs.ollama.com/capabilities/thinking

<!-- gh-comment-id:4107185846 --> @OdinVex commented on GitHub (Mar 22, 2026): `for t in low medium high` ... `"reasoning_effort":"'$t'"` <-- Looks like you're using `reasoning_effort`, not `think`. The output being different (much better with `high`) is exactly what I'm talking about, too. `think` only controls what is seen for trace, not whether the response has higher reasoning. Think and Reasoning Effort are not the same. Reasoning Effort: https://developers.openai.com/api/docs/guides/reasoning Think: https://docs.ollama.com/capabilities/thinking
GiteaMirror commented 2026-05-10 00:17:45 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

for t in low medium high ... "reasoning_effort":"'$t'" <-- Looks like you're using reasoning_effort, not think.

reasoning_effort is only supported in the OpenAI API. If I were to use the Ollama API, then the API call would use think.

Internally, they are the same. The some amount of reasoning happens for think:true as for reasoning_effort:medium. gpt-oss has additional levels of thinking that other models do not, hence low and high are only supported in gpt-oss.

<!-- gh-comment-id:4107190491 --> @rick-github commented on GitHub (Mar 22, 2026): > for t in low medium high ... "reasoning_effort":"'$t'" <-- Looks like you're using reasoning_effort, not think. `reasoning_effort` is only supported in the OpenAI API. If I were to use the Ollama API, then the API call would use `think`. Internally, they are the same. The some amount of reasoning happens for `think:true` as for `reasoning_effort:medium`. gpt-oss has additional levels of thinking that other models do not, hence `low` and `high` are only supported in gpt-oss.
GiteaMirror commented 2026-05-10 00:17:46 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

"Thinking-capable models emit a thinking field that separates their reasoning trace from the final answer." ... "GPT-OSS instead expects one of low, medium, or high to tune the trace length." I don't care about think trace length, I'm trying to affect Reasoning Effort, which (GPT-OSS being a reasoning model from OpenAI's introduction of it) should support via reasoning_effort.

<!-- gh-comment-id:4107197533 --> @OdinVex commented on GitHub (Mar 22, 2026): "Thinking-capable models emit a thinking field that separates their reasoning trace from the final answer." ... "GPT-OSS instead expects one of low, medium, or high to tune the trace length." I don't care about `think` trace length, I'm trying to affect Reasoning Effort, which (GPT-OSS being a reasoning model from OpenAI's introduction of it) should support via `reasoning_effort`.
GiteaMirror commented 2026-05-10 00:17:47 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

should support via reasoning_effort.

Which it does. I just showed you how to use it.

<!-- gh-comment-id:4107198917 --> @rick-github commented on GitHub (Mar 22, 2026): > should support via reasoning_effort. Which it does. I just showed you how to use it.
GiteaMirror commented 2026-05-10 00:17:49 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

for t in low medium high ... "reasoning_effort":"'$t'" <-- Looks like you're using reasoning_effort, not think.

<!-- gh-comment-id:4107200328 --> @OdinVex commented on GitHub (Mar 22, 2026): > `for t in low medium high` ... `"reasoning_effort":"'$t'"` <-- Looks like you're using `reasoning_effort`, not `think`.
GiteaMirror commented 2026-05-10 00:17:51 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

Yes, because the one line script is calling the OpenAI API.

<!-- gh-comment-id:4107201384 --> @rick-github commented on GitHub (Mar 22, 2026): Yes, because the one line script is calling the OpenAI API.
GiteaMirror commented 2026-05-10 00:17:54 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

reasoning_effort != think

It's alright to say you can't/won't fix it, I can privately fork and fix it for now.

xhigh is also available, by the way.

<!-- gh-comment-id:4107215729 --> @OdinVex commented on GitHub (Mar 22, 2026): `reasoning_effort` != `think` It's alright to say you can't/won't fix it, I can privately fork and fix it for now. `xhigh` is also available, by the way.
GiteaMirror commented 2026-05-10 00:17:55 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

I look forward to the PR.

<!-- gh-comment-id:4107219005 --> @rick-github commented on GitHub (Mar 22, 2026): I look forward to the PR.
GiteaMirror commented 2026-05-10 00:17:57 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 22, 2026):

I look forward to the PR.

What for? I'm taking a stop-gap approach until an in-house alternative is put together that actually supports sharded GGUFs and fixes a number of issues frustrating me and others with Ollama. Ollama just seems to be interested in pressing Cloud rather than this.

<!-- gh-comment-id:4107223060 --> @OdinVex commented on GitHub (Mar 22, 2026): > I look forward to the PR. What for? I'm taking a stop-gap approach until an in-house alternative is put together that actually supports sharded GGUFs and fixes a number of issues frustrating me and others with Ollama. Ollama just seems to be interested in pressing Cloud rather than this.
GiteaMirror commented 2026-05-10 00:17:58 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 22, 2026):

What for?

Because I'd like to know what you think the actual problem is. Internally, think and reasoning_effort are the same. Here's the code:

22c2bdbd8a/openai/openai.go (L611-L644)

reasoning_effort from the OpenAI compatible endpoint is converted to a Think value that is passed through the ChatRequest interface to the model.

<!-- gh-comment-id:4107240844 --> @rick-github commented on GitHub (Mar 22, 2026): > What for? Because I'd like to know what you think the actual problem is. Internally, `think` and `reasoning_effort` are the same. Here's the code: https://github.com/ollama/ollama/blob/22c2bdbd8add79c6fca2eb2538ab36fcd4a9ca48/openai/openai.go#L611-L644 `reasoning_effort` from the OpenAI compatible endpoint is converted to a `Think` value that is passed through the ChatRequest interface to the model.
GiteaMirror commented 2026-05-10 00:17:59 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 23, 2026):

The actual problem is the misunderstanding around reasoning_effort and think. They're not the same thing. think, is for the thought process output (the chain-of-thought output and how detailed/long/terse it can be, on/off for some models, other models support varying depths of output). reasoning_effort influences how many tokens are used in generating a response, such as considering alternatives or more computation before generating an answer.

https://openai.com/index/introducing-gpt-oss/ for the values GPT-OSS supports. GPT-OSS supports low, medium, high.
https://developers.openai.com/api/docs/guides/reasoning for all potential values for reasoning_effort such as minimal, xhigh, etc. Not all models support reasoning_effort and not all models support certain values.

<!-- gh-comment-id:4107309541 --> @OdinVex commented on GitHub (Mar 23, 2026): The actual problem is the misunderstanding around `reasoning_effort` and `think`. They're *not* the same thing. `think`, is for the thought process *output* (the chain-of-thought output and how detailed/long/terse it can be, on/off for some models, other models support varying depths of output). `reasoning_effort` influences how many tokens are used in generating a response, such as considering alternatives or more computation before generating an answer. https://openai.com/index/introducing-gpt-oss/ for the values GPT-OSS supports. GPT-OSS supports low, medium, high. https://developers.openai.com/api/docs/guides/reasoning for all potential values for reasoning_effort such as `minimal`, `xhigh`, etc. Not all models support `reasoning_effort` and not all models support certain values.
GiteaMirror commented 2026-05-10 00:18:01 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 23, 2026):

I look forward to the PR.

<!-- gh-comment-id:4107320142 --> @rick-github commented on GitHub (Mar 23, 2026): I look forward to the PR.
GiteaMirror commented 2026-05-10 00:18:01 -05:00
Author
Owner
Copy Link

@OdinVex commented on GitHub (Mar 23, 2026):

https://openai.com/index/introducing-gpt-oss/ "Similar to the OpenAI o-series reasoning models in the API, the two open-weight models support three reasoning efforts—low, medium, and high—which trade off latency vs. performance. Developers can easily set the reasoning effort with one sentence in the system message. "

https://developers.openai.com/api/docs/guides/reasoning "Supported values are model-dependent and can include none, minimal, low, medium, high, and xhigh. Lower effort favors speed and lower token usage, while higher effort favors more complete reasoning. Defaults are also model-dependent rather than universal. For example, gpt-5.4 defaults to none, while older GPT-5 models default to medium."

<!-- gh-comment-id:4107339977 --> @OdinVex commented on GitHub (Mar 23, 2026): https://openai.com/index/introducing-gpt-oss/ "Similar to the OpenAI o-series reasoning models in the API, the two open-weight models support three reasoning efforts—low, medium, and high—which trade off latency vs. performance. Developers can easily set the reasoning effort with one sentence in the system message. " https://developers.openai.com/api/docs/guides/reasoning "Supported values are model-dependent and can include none, minimal, low, medium, high, and xhigh. Lower effort favors speed and lower token usage, while higher effort favors more complete reasoning. Defaults are also model-dependent rather than universal. For example, gpt-5.4 defaults to none, while older GPT-5 models default to medium."
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label app feature request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#85482
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?