This page compares a near-flagship model with a fast mid-tier model on one job: turning a long customer call transcript into themes and action items. It tests them through their APIs and shows when the cheaper model is enough.
Oct 13, 2026 · 10 min read
Messy calls and high-stakes summaries favor GPT-6.1 Sol. Clean, labelled and reviewed transcripts at volume favor Gemini 3.8 Flash.
On AA-LCR, a test that asks models to connect information across document sets of around 100,000 tokens, Gemini 3.8 Flash scored 84% and GPT-6.1 Sol 83% at medium effort1. That is a tie for practical purposes. On the broader tests Sol is ahead. It led the composite Intelligence Index 48 to 40 and AA-Briefcase 1,361 to 1,1811.
The framing needs one correction. OpenAI positions GPT-6 Astra as its flagship, and GPT-6.1 Sol is the lower-cost, near-flagship option2. So this is a near-flagship model against a fast mid-tier one, and the useful question is whether the cheaper model is enough.
A staged workflow follows. Use Gemini for a first pass that pulls out topics, decisions, explicit actions and supporting transcript lines. Use Sol for the final summary when the result will influence an escalation, a renewal or an executive decision. That split is a judgment call and not a measured rule.
You summarize many clean, labelled calls and a person reads each draft. Flash is nearly tied with Sol on the long-context test and costs less per token.
A missed objection or a tentative commitment can change a renewal. Sol led the broader reasoning tests, so it is the safer choice when the summary drives a decision.
Your calls vary from tidy to chaotic. Run a blind test on a mix of both and see where the cheaper model stops being enough.
You process calls in bulk. Let Flash do the first pass and send only low-confidence or high-consequence calls to Sol.
This page compares the two models through their API in one neutral setup, and leaves out the meeting apps and plug-ins that sit around a model.
A call summary here means a long text transcript turned into themes and action items. The parts that matter are grouping scattered comments into themes, separating customer statements from vendor statements, attributing decisions and actions to the right speaker, telling firm commitments from suggestions, and keeping deadlines, conditions and open questions. We start from the official documentation and then the Artificial Analysis comparison of the two exact models at medium effort1, 7.
We left tools out of the specs table on purpose. Recording, transcription, CRM plug-ins and file upload belong to the app around the model, so the same model behaves differently in a meeting tool, a chat product, the API or a workspace. Judging those here would compare wrappers and say little about summarizing.
The model facts that shape a transcript job. Tool and file features are left out, since those depend on the app around the model.
Figures from OpenAI and Google documentation, checked October 2026. A token is the small chunk of text a model reads and bills by, and 1M means one million of them. The two vendors tokenize differently, so treat any cross-model cost comparison as directional.
The answer changes by job and by how clean the transcript is. This is the main analysis, with the edge on each part of summarizing a call.
Sol looks stronger at interpreting what a conversation means. Flash looks just as viable at finding material in a long, clean transcript and expressing it in a set structure. Transcript length alone is not a reason to pay for Sol, and ambiguity, reputation risk and expected editing time are better reasons.
A useful test is repetitive. Same transcripts, same prompt, same reasoning level, same output schema, then judge each summary on what a reviewer would change.
Choose three to five real transcripts. Include a clean call, a call full of interruptions, a call with conflicting commitments, and one where important information comes early and is revised later.
One prompt with the same schema and the same evidence rule for both models. Neither model gets a richer version, and any change you make mid-test applies to both.
Use the same transcript text and the same reasoning level, and run both in the place your team will work. Chat products add system prompts and tools, so API and chat results can differ.
Check whether each summary found every material theme, kept minority and negative feedback, gave each statement to the right speaker, and separated decisions from proposals and open questions. Check that each action has evidence or says no owner or date was stated, that the schema held, and how much hand editing was needed. For customer-facing work, hide the model names and use at least two reviewers.
No public benchmark gives both exact models the same customer calls, so the evidence is a mix of related tests and vendor statements. Here is what each source helps judge.
No source reviewed provides an exact-model, like-for-like test of customer-call themes, speaker attribution and action extraction. Claims about which model drops a tentative commitment or mixes up two speakers are workflow judgments to validate internally.
Both prompts ask for evidence behind every item and a fixed schema. They differ in how the instructions are arranged and how much reasoning to use.
GPT-6.1 Sol does well with a lean instruction hierarchy and an explicit evidence rule. Keep instructions free of repeats and conflicts, since OpenAI's structured-output support can enforce the action schema3. Start at medium reasoning and raise it only if your test shows a gain on ambiguous calls.
Gemini 3.8 Flash does well when the whole transcript comes first and the question and schema come after it. Google recommends placing long context before the specific instructions and ending with a clear transition9. Medium thinking is the natural start for routine calls, with high kept for difficult synthesis.
A GPT-6.1 Sol prompt: an evidence rule for every item
Summarize the transcript into: themes, decisions, risks,
unresolved questions and action items.
For every item, include the speaker, the timestamp or turn ID,
and a short evidence excerpt.
Never infer an owner or deadline. Return null if unstated.
Merge repeated points, but preserve disagreement and later
corrections.A Gemini 3.8 Flash prompt: transcript first and the schema after
[TRANSCRIPT]
Based only on the transcript above, return the required JSON
schema. Keep customer and company viewpoints separate.
A commitment counts as an action only when the speaker indicates
intent to do it. Include turn IDs as evidence. Mark inferred
themes separately from explicitly stated themes.Neither model is safe to trust blindly on a customer call. The useful question is where each one adds cleanup work, and what to change in the prompt or the workflow.
One question first: would a missed nuance change a customer decision. Then follow the branch that matches the call and the volume.
Use this as a starting point and test on your own calls before you commit
If a missed nuance could change a renewal, an escalation, a compliance decision or an executive review, choose GPT-6.1 Sol and require evidence for every item1. If several calls must be summarized together, watch the 272,000-token threshold where the Sol price rises to $4 in and $15 out per million tokens3.
If the output is an internal draft that a person reviews, choose Gemini 3.8 Flash. With clear speaker labels and a stable schema it should usually be enough, and at high volume its lower prices add up quickly4. For a clean long transcript where theme recall matters most, start with Gemini, since the long-context test does not show a Sol advantage1.
If the calls hold contradictions, hidden objections or unclear ownership, choose Sol. If cost matters most and a person still reviews, run Gemini first and send only the flagged calls to Sol.
If one person summarizes a few calls a month with a single model, Playgram is not the right buy. Its value is in comparing models and sharing context across a team.
GPT-6.1 Sol is the safer single model for call summaries. Gemini 3.8 Flash is the better economic default for standard, human-reviewed ones, and routing between them is often the practical answer.
The limits are real. No public exact-model test measures theme recall, speaker attribution or action accuracy on customer calls. The independent long-context result is nearly tied while the broader reasoning tests favor Sol. Vendor benchmarks use different prompts, effort levels and comparison models, and Gemini's current price is introductory and changes on January 1, 20274, 8.
The safest final step is to test the shape of your own calls with your own schema. A fair test needs the same setup for both models: the same transcript text, the same prompt and the same place to run them, so the result reflects the models and not the tool around them. In practice that is harder than it sounds, since most teams end up running one model in one app and the other in a different one, on two separate subscriptions, which tilts the comparison before the first summary comes back. The cleaner the setup, the more the difference you see is really GPT-6.1 Sol vs Gemini 3.8 Flash, and which one happened to be easier to reach that day matters less.
Upgrade as needed, and only pay for what you actually use
Save ~17% with the annual plan
Pro
Perfect for small and medium teams
Unlimited users & infinite memory
Multi-LLM chats
Granular access control to models
EU data residency
Ultra
Best for large, growing teams
Unlimited users & infinite memory
Multi-LLM chats
Unlimited use of DeepSeek V4 Flash
Granular access control to models
Choose US or EU data residency
Enterprise
Get in touch
For organizations with advanced needs
Unlimited users & SSO
Priority Support
Unlimited use of DeepSeek V4 Flash
Granular access control to models
Choose US or EU data residency
30-days money back guarantee