Batchrate @batchrate

LLM inference for jobs that can wait. OpenAI-compatible Batch API, results within hours. Big LLM jobs. Small bills. batchrate.ai Joined October 2026

Tweets

59
Followers

11
Following

50
Likes

40

Batchrate @batchrate

a day ago

Classifying with a chat model means paying for output tokens and parsing JSON that sometimes breaks. Decisions always answers from your choices and returns probabilities you can threshold on. Same /v1/batches API: url "/v1/decisions", model "batchrate-decide". batchrate.ai/docs?ref=x-dec…

0 0 1 45 0

View Details

Batchrate @batchrate

2 days ago

@forestchav That's not very nice. Its X's ads that spammed you, not me.

0 0 0 32 0

View Details

Batchrate @batchrate

4 days ago

Running gpt-oss batch jobs on Doubleword? Those models stop accepting requests after Oct 30. You can keep the same model and the same code. Change base_url, the key and the model name, and you're done. 120b is ~25% less than Doubleword batch. batchrate.ai/migrate?ref=x

2 12 153 19K 9

View Details

Batchrate @batchrate

2 days ago

Rough math on batch pricing: • Qwen3.6 35B: about 25% less than Doubleword • 55-62% below OpenAI batch gpt-4.1-mini / gpt-5-mini (price only; different models) Test it on your own data first.

0 0 0 14 0

View Details

Batchrate @batchrate

2 days ago

Your 24h batch job just became urgent? Same model, same API, one flag. Rush finishes it within the hour, or you pay standard. batchrate.ai/docs?ref=x-rus…

0 0 0 11 0

View Details

Batchrate @batchrate

3 days ago

@AntoniusBlcbai 👀

2 0 0 77 0

View Details

Batchrate @batchrate

3 days ago

Switching batch providers is 3 lines: base_url="api.batchrate.ai/v1" api_key=os.environ["BATCHRATE_API_KEY"] model="gpt-oss-120b" Your JSONL, polling and parsing don't change. gpt-oss-120b and 20b are Apache 2.0 and made in the US by OpenAI. batchrate.ai/docs?ref=x

0 0 1 33 0

View Details

Batchrate @batchrate

3 days ago

@AICreditRadar Good list. One more: classification jobs (routing, tagging, moderation) often don't need free text at all. Batchrate Decisions returns one label plus probabilities for every choice at $0.02/M input, output free: batchrate.ai/?ref=x

0 0 1 4 0

View Details

Batchrate @batchrate

3 days ago

@dianne_alter Design seems like a fair miss. Where open models tend to hold up is the unglamorous work: tagging, extraction, summaries and eval grading, where you can check outputs against a labeled set. Run those as batch jobs and the price gap you measured gets wider again.

0 0 1 9 0

View Details

Batchrate @batchrate

3 days ago

@cwolferesearch Great list. On the noise side, most of the fixes (multiple judge samples, swapped orderings, bootstrapped error bars) multiply the number of calls, and none of them need a live answer. Running the eval suite through a batch API keeps those extra samples affordable.

0 0 0 4 0

View Details

Batchrate @batchrate

3 days ago

@solsona The order-swap check is cheap to make routine: run every pairwise judgment twice with A and B flipped, and only count verdicts that agree. Doubling judge calls hurts a lot less when the suite runs as an overnight batch job instead of inside CI.

0 0 0 4 0

View Details

Batchrate @batchrate

3 days ago

@Al_Grigor Good one to include. Eval suites are close to the ideal batch workload: fixed inputs, nobody waiting, and you rerun them on every prompt change. Worth adding open-weight judges in batch too, e.g. gpt-oss-120b at $0.06/M input: batchrate.ai/?ref=x

0 0 0 1 0

View Details

Batchrate @batchrate

4 days ago

@anupamrjp batchrate.ai

0 0 0 6 0

View Details

Batchrate @batchrate

4 days ago

Gemma 4 26B A4B is now on Batchrate. $0.0375 in / $0.1275 out per 1M tokens, about 25% less than Doubleword batch. On Doubleword? It stops taking requests after Oct 30. Keep the same model: swap base_url and key. batchrate.ai/migrate?ref=x

0 0 1 19 0

View Details

Batchrate @batchrate

4 days ago

New on Batchrate: Decisions. Give each row a prompt and a list of choices. Get back one label plus probabilities for every choice, calibrated on labeled test sets. $0.02 per million input tokens. Output is free. batchrate.ai/docs?ref=x-dec…

0 0 1 24 0

View Details

Batchrate @batchrate

4 days ago

New on Batchrate: Rush. Batch results within the hour at 1.5x standard. Miss the hour and you pay the standard price. batchrate.ai/pricing?ref=x-…

0 0 2 46 0

View Details

Batchrate @batchrate

4 days ago

gpt-oss-120b and gpt-oss-20b are now on Batchrate. Batch pricing per 1M tokens: - 120b: $0.06 in / $0.225 out - 20b: $0.015 in / $0.0525 out OpenAI-compatible Batches API, signed webhooks, and 10M free tokens to start. batchrate.ai/migrate?ref=x

0 0 1 29 0

View Details

Batchrate @batchrate

5 days ago

@emankhoubian @nikships Agreed, a wrong transcript looks finished. At that volume, scoring a small human-checked sample from each run catches drift early. For backlog audio that can wait, batch Whisper Large v3 Turbo is $0.0006/min on our side, so side-by-side model checks stay cheap.

0 0 0 51 0

View Details

Batchrate @batchrate

5 days ago

@aimorebou Cost per task, every time. Latency tolerance belongs in that test too: the same task through a batch API usually costs a lot less than sync. A column for "can this wait a few hours?" changes which model wins more often than you'd think.