# AI models comparison 2026: text, code, image, video, voice

URL: https://infro.io/blog/ai-models-comparison-2026
Author: INFRO Editorial (https://infro.io/blog/editorial)
Publisher: INFRO
Published: 2026-10-19
Last modified: 2026-10-19
Last reviewed: 2026-10-19 (next review due 2026-11-18)
Topic: Evaluation and models
Format: Comparison

> On vendor pages read on 19 October 2026, frontier text models list from $2 to $10 per million input tokens and $6 to $50 output, small models from $0.10 input, image models from $0.02 per image, and video from $0.02 to $0.60 per second. Rates are billed in different units and change on dated schedules, so compare models on cost per successful task measured on your own workload, not on the rate card.

An AI models comparison is only useful if it compares like with like and says when it was read. This page puts the main models of each modality — frontier text, coding, small and fast, image, video and voice — in one table each, with the vendor's own list rate, context window and limits as published on 19 October 2026. Each table links to the deeper comparison for that tier, and the last section is the part no table can do for you: how to compare AI models on your own workload.

Every rate below is the model vendor's list rate: what an account without a negotiated agreement is billed by that vendor. If your organization has negotiated rates, those replace it. For the full catalog of models with reference rates, searchable and sortable, see the [model directory](https://infro.io/models).

## How to read these tables

Three rules keep a comparison table honest.

- **Units differ by modality.** Text is billed per million tokens, most image models per image or per megapixel, video per second of output, and speech per character or per hour. Do not set a per-token number beside a per-image number.
- **Rates move on schedules.** Several rates below change by prompt length, time of day or date. The tables give the base rate and name the condition; the deeper comparisons work through the arithmetic.
- **"Not stated" means not stated.** Where the vendor page read on 19 October did not show a fact, the table says so rather than carrying an older figure forward.

## Frontier text and reasoning models

These are the models each vendor positions as its most capable for reasoning and long, complex work.

| Vendor | Model | Input / output per 1M | Context | Max output | Notes |
| --- | --- | --- | --- | --- | --- |
| OpenAI | GPT-6 Astra | $10 / $50[1] | 1.05M[2] | 128K[2] | Above 272K input: $20 / $75 for the request[1] |
| OpenAI | GPT-6.1 Sol | $2 / $10[1] | 1.05M[2] | 128K[2] | Above 272K input: $4 / $15[1] |
| Anthropic | Claude Fable 5.1 | $10 / $50[3] | 1M[4] | 128K[4] | Full 1M window at the standard rate[3] |
| Anthropic | Claude Opus 5.5 | $4 / $20[3] | 1M[4] | 128K[4] | Cache hits at $0.20[3] |
| Anthropic | Claude Sonnet 5.5 | $2 / $10[3] | 1M[4] | 128K[4] | Cache hits at $0.10[3] |
| Google | Gemini 3.1 Pro | $2 / $12[5] | 1,048,576[6] | 65,536[6] | Preview; above 200K: $4 / $18[5][6] |
| xAI | Grok 4.7 | $2 / $6[7] | 500K[7] | Not stated | At 200K or more, $4 / $12 for the request[7] |
| Moonshot AI | Kimi K3 | $3 / $15[8] | 1,048,576[8] | Not stated | Cache writes $3 (5 min) or $6 (1 h)[8] |
| DeepSeek | DeepSeek V4 Pro | $1.32 / $3.96[9] | 1M[9] | 384K[9] | Peak rate; off-peak is half[9] |
| Z.ai | GLM-5.3 | $1.40 / $4.40[10] | Not stated | Not stated | Cached input $0.26[10] |
| Alibaba Cloud | Qwen3.8 Max | $2 / $6[11] | Up to 1M priced[11] | Not stated | International (Singapore) deployment[11] |

Table: Frontier text models, vendor list rates per million tokens, read 19 October 2026.

Two lines in that table matter more than the headline rate. GPT-6 Astra and Claude Fable 5.1 list the same $10 and $50, but OpenAI reprices a request above 272K input tokens while Anthropic bills the whole 1M window at the standard rate[1][3]. And Gemini 3.1 Pro is still published under a preview model code[6], a lifecycle fact that belongs in any production decision. The [frontier models comparison](https://infro.io/blog/frontier-models-compared-2026) covers caching, batch and regional uplifts for the top four.

## Coding models

Coding agents resend large prompts many times per task, so context, caching and prompt-length tiers shape the bill more than in chat.

| Vendor | Model | Input / output per 1M | Cached input | Context | Notes |
| --- | --- | --- | --- | --- | --- |
| Anthropic | Claude Opus 5.5 | $4 / $20[3] | $0.20[3] | 1M[4] | Anthropic positions it for long-running agentic coding[4] |
| Moonshot AI | Kimi K2.7 Code | $0.95 / $4[8] | $0.19[8] | 262,144[8] | A highspeed variant lists $1.90 / $8[8] |
| MiniMax | MiniMax M3 | $0.30 / $1.20[12] | $0.06[12] | Not re-checked | Up to 512K input; $0.60 / $2.40 above[12] |
| xAI | Grok Build 0.1 | $1 / $2[7] | $0.20[7] | 256K[7] | Doubles at 200K or more[7] |
| Mistral AI | Mistral Medium 3.5 | Not read | Not read | Not stated | Positioned for agentic and coding use[13] |

Table: Coding models, read 19 October 2026.

One change since the [coding models comparison](https://infro.io/blog/coding-models-compared-2026) was written: on 19 October Alibaba Cloud's international pricing page no longer listed `qwen3-coder-plus`; the only Qwen3-Coder model on it was `qwen3-coder-flash`[11]. If you built a cost model on the coder-plus tiers, check the model is still offered in your region before relying on it.

## Small and fast models

The tier that carries most production traffic by volume: classification, extraction, routing and short summaries.

| Vendor | Model | Input / output per 1M | Context | Rate condition |
| --- | --- | --- | --- | --- |
| OpenAI | GPT-6 Luna | $0.10 / $0.50[1] | 1.05M[2] | $0.20 / $0.75 above 272K input[1] |
| Anthropic | Claude Haiku 5.5 | $0.10 / $0.50[3] | 1M[4] | $0.50 / $2.50 for prompts over 100,000 tokens[3] |
| Anthropic | Claude Haiku 4.5 | $1 / $5[3] | Not re-checked | Now listed as a legacy model[4] |
| Google | Gemini 3.8 Flash | $0.75 / $3.75[5] | 1,048,576[14] | Doubles to $1.50 / $7.50 on 1 January 2027[5] |
| DeepSeek | DeepSeek Flash (V4.1) | $0.30 / $1.20[9] | 1M[9] | Peak rate; $0.15 / $0.60 off-peak[9] |
| Z.ai | GLM-5.3 Flash | $0.15 / $0.50[10] | Not stated | — |
| Alibaba Cloud | Qwen3.8 Flash | $0.15 / $0.47[11] | Not stated | International deployment[11] |

Table: Small and fast models, read 19 October 2026.

Claude Haiku 5.5 is not in the [small and fast models comparison](https://infro.io/blog/small-fast-models-compared-2026). Its rate depends on prompt length: a request over 100,000 tokens pays the higher rate on the whole request, cache reads and writes included[3]. DeepSeek's peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays, excluding Chinese public holidays[9].

## Open-weight models

Open weights separate the lab that trained a model from the host that runs it, so the rate depends on who hosts it. The facts that travel with the model are its license and its limits.

| Lab | Model | License, as stated | Context | First-party API rate per 1M |
| --- | --- | --- | --- | --- |
| OpenAI | gpt-oss-120b | Apache 2.0[15] | 131,072[15] | Run on your own hardware or a host; fits one H100[15] |
| Mistral AI | Mistral Large 3 | Apache 2.0[13] | Not stated | Not read |
| Mistral AI | Mistral Small 4 | Apache 2.0[13] | Not stated | Not read |
| DeepSeek | DeepSeek V4.1 Flash | Read the model card | 1M on DeepSeek's API[9] | $0.30 / $1.20 peak[9] |
| MiniMax | MiniMax M3 | Read the model card | Not re-checked | $0.30 / $1.20[12] |

Table: Open-weight models, read 19 October 2026. Other hosts set their own rates and limits.

For the Chinese labs' families, where each publishes weights, and the hosting and compliance questions, see the [open-weight models guide](https://infro.io/blog/chinese-open-weight-models-guide-2026).

## Image generation models

Image models bill in three different units, so the column that matters is the unit.

| Vendor | Model | Billing unit | List rate | Notes |
| --- | --- | --- | --- | --- |
| OpenAI | GPT Image 2.5 Sunburst | Per token | $30 per 1M image output tokens; $5 text input, $8 image input[16] | No per-image estimate published for 2.5[16] |
| Google | Nano Banana 2 | Per image, by resolution | $0.045 (0.5K), $0.067 (1K), $0.101 (2K), $0.151 (4K)[5] | — |
| Google | Nano Banana Pro | Per image, by resolution | $0.134 (1K or 2K), $0.24 (4K)[5] | — |
| xAI | Grok Imagine | Per image, by variant | $0.02, $0.04 (2.0), $0.05 (quality)[7] | — |
| Black Forest Labs | FLUX.2 family | Per megapixel | Not shown when read[17] | Rates load in the page's calculator |

Table: Image generation models, read 19 October 2026.

Retirements belong in this table too. OpenAI shuts down `gpt-image-1` on 23 October 2026 and `gpt-image-1.5` and `gpt-image-1-mini` on 1 December 2026, naming the GPT Image 2.5 models as replacements[18]. The [image models comparison](https://infro.io/blog/ai-image-models-compared-2026) covers sizes, reference images and how results come back.

## Video generation models

Video is billed per second of output, and resolution usually moves the rate.

| Vendor | Model | Per second, list rate | Notes |
| --- | --- | --- | --- |
| Google | Veo 3.1 | $0.40 (720p, 1080p), $0.60 (4K)[5] | Gemini API preview ids listed for shutdown on 22 October 2026[19] |
| Google | Veo 3.1 Fast / Lite | $0.10–$0.30 / $0.05–$0.08[5] | Same shutdown date[19] |
| Google | Gemini Omni Flash | $17.50 per 1M video output tokens, about $0.10 a second at 720p[5] | Google names `gemini-omni-1.1-flash` as the Veo replacement[19] |
| xAI | Grok Imagine Video 1.5 | $0.08; Lite $0.02[7] | — |
| MiniMax | MiniMax H3 | $0.08 (768P), $0.13 (2K)[12] | H3 Max $0.05 at 480P[12] |
| Alibaba Cloud | Wan 2.7 | $0.10 (720P), $0.15 (1080P)[11] | — |
| Alibaba Cloud | Wan 3.0 Prime | $0.068 (480P), $0.14 (720P), $0.28 (1080P)[11] | — |

Table: Video generation models, read 19 October 2026.

The Veo 3.1 line is the reason a video comparison needs a date: Google's deprecation page lists 22 October 2026, three days after this page was read, as the shutdown date for the three Veo 3.1 preview model ids, and notes that listed dates are the earliest possible ones[19]. Kling, Seedance and Runway are covered in the [video models comparison](https://infro.io/blog/ai-video-models-compared-2026); their rates were not re-read for this page.

## Voice: speech and transcription models

| Vendor | Model | Task | List rate | Notes |
| --- | --- | --- | --- | --- |
| ElevenLabs | Eleven v4, Eleven v3 | Text to speech | $0.08 per 1K characters[20] | Flash and Turbo $0.04 per 1K[20] |
| ElevenLabs | Scribe v2 | Transcription | $0.22 per hour[20] | — |
| OpenAI | tts-1 / tts-1-hd | Text to speech | $15 / $30 per 1M characters[1] | Shut down 6 January 2027[18] |
| OpenAI | gpt-transcribe | Transcription | $0.0045 per minute[1] | Named replacement for `whisper-1`[18] |
| Google | Gemini 3.8 Flash TTS | Text to speech | $0.50 text in, $9 audio out per 1M tokens[5] | Doubles on 1 January 2027[5] |
| MiniMax | Speech 2.8 HD / Turbo | Text to speech | $100 / $60 per 1M characters[12] | — |
| xAI | Grok text to speech, speech to text | Both | $15 per 1M characters; $0.10 per hour REST[7] | Streaming transcription $0.20 per hour[7] |

Table: Speech and transcription models, read 19 October 2026.

Speech is the modality where the billing unit varies most within one vendor: OpenAI prices `tts-1` per character and `gpt-4o-mini-tts` per token[1]. Convert to cost per minute of finished audio at your own average speaking rate before comparing.

## How to compare AI models on your own workload

The tables tell you which models can run a task and what the provider will bill per unit. They cannot tell you which model does the task well enough, or what a successful result costs. That needs your own data.

1. **Filter on hard constraints first.** Context window, output limit, input modalities, region, lifecycle status (preview, retiring) and license. This usually removes half of any table above.
2. **Build an evaluation set from real traffic.** Fifty to a few hundred real inputs per task type, each with a check a machine or a reviewer can apply: schema valid, tests pass, label matches, image accepted. The guide to [evaluating LLMs on real workloads](https://infro.io/blog/evaluate-llms-real-workloads) covers how to build one.
3. **Run every shortlisted model on the same inputs, under the same conditions.** Same prompt, same tools, same effort level, the stable prefix cached the way you would ship it.
4. **Record estimated provider cost and outcome per request.** Include retries and failed attempts; they are billed too.
5. **Compute cost per successful task.** Total estimated cost divided by the number of tasks that passed. The companion piece on [cost per successful task](https://infro.io/blog/cost-per-successful-task) goes deeper.
6. **Re-run on a dated trigger.** A rate change, a new model or a retirement on the calendar is a reason to repeat the run, not to trust last quarter's result.

> **Example: Cost per successful task across two tiers**
> Illustrative arithmetic. An extraction task averages 4,000 input and 500 output tokens. A small model at $0.10 / $0.50 per million costs $0.00065 per request; a mid-tier model at $2 / $10 costs $0.013. On 1,000 tasks the small model passes 880 and the mid-tier model 980. If the small model's 120 failures are retried on the mid-tier model and pass, the cascade costs $0.65 + 120 × $0.013 = $2.21 for 1,000 successes, against $13.00 ÷ 980 × 1,000 ≈ $13.27 for the mid-tier model alone. Your pass rates will differ; that is why they have to be measured.

## Comparing models through INFRO

INFRO does not score outputs; your evaluation harness does. What it gives a comparison is one OpenAI-compatible endpoint in front of the provider accounts you connect, so every candidate is called the same way, and a record of each request with its estimated provider cost. Your providers bill your accounts directly at your rates; INFRO's subscription is separate and nothing it sells is priced per token.

- **Run candidates side by side.** The console Playground runs text, image, video, speech and transcription on one of your connections or through a gateway key, and copies the call as cURL.
- **Estimate from your rates.** Each response carries `usage.cost`, an estimate of what your provider will bill, and `price_source`. Enter negotiated rates under Prices ([billing](https://infro.io/docs/org/billing)) so the comparison ranks rates you actually pay, not list rates.
- **Group the run.** `GET /v1/usage?group_by=model` returns requests, estimated cost, unpriced requests, error rate and latency percentiles per model ([usage analytics](https://infro.io/docs/observability/analytics)).
- **Ship the decision as configuration.** A [route](https://infro.io/docs/routes) names the chosen model first and a tested alternative second, so a retirement like Veo 3.1's is a console edit, not a code change ([routing](https://infro.io/docs/features/routing)).

The catalog carries one rate per model, so a request above a vendor's long-prompt threshold or in DeepSeek's off-peak window is estimated at the base rate; check those against the provider's own usage data.

## Questions

### Why do AI model rates look so hard to compare?

Because they are billed in different units. Text is per million tokens, most image models per image or per megapixel, GPT Image per token, video per second and speech per character or per hour. Convert each to the unit your product counts — a resolved ticket, an accepted image, a finished clip — before putting two models side by side.

### Is the list rate what my company will actually pay?

Only if you have no agreement with the provider. A negotiated rate on your own account replaces the list rate, and some vendors change rates by time of day, prompt length or date. Use list rates to shortlist, and your own rates and invoices to decide.

### How often do these rates change?

Often enough that a comparison table needs a date. On 19 October 2026 Anthropic listed Claude Haiku 5.5, which the late-September tier comparisons do not cover, and Google's deprecation page listed 22 October 2026 for the Veo 3.1 preview models. This page is reviewed every 30 days.

## Sources

1. Pricing — OpenAI. https://developers.openai.com/api/docs/pricing (accessed 2026-10-19)
2. Models — OpenAI. https://developers.openai.com/api/docs/models (accessed 2026-10-19)
3. Pricing — Anthropic. https://platform.claude.com/docs/en/about-claude/pricing (accessed 2026-10-19)
4. Models overview — Anthropic. https://platform.claude.com/docs/en/about-claude/models/overview (accessed 2026-10-19)
5. Gemini Developer API pricing — Google. https://ai.google.dev/gemini-api/docs/pricing (accessed 2026-10-19)
6. Gemini 3.1 Pro Preview — Google. https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview (accessed 2026-10-19)
7. Models and pricing — xAI. https://docs.x.ai/docs/models (accessed 2026-10-19)
8. Chat model pricing — Moonshot AI. https://platform.kimi.ai/docs/pricing/chat (accessed 2026-10-19)
9. Models & pricing — DeepSeek. https://api-docs.deepseek.com/quick_start/pricing/ (accessed 2026-10-19)
10. Pricing — Z.ai. https://docs.z.ai/guides/overview/pricing (accessed 2026-10-19)
11. Model pricing — Alibaba Cloud Model Studio. https://www.alibabacloud.com/help/en/model-studio/model-pricing (accessed 2026-10-19)
12. Pay-as-you-go pricing — MiniMax. https://platform.minimax.io/docs/guides/pricing-paygo (accessed 2026-10-19)
13. Models overview — Mistral AI. https://docs.mistral.ai/getting-started/models/models_overview/ (accessed 2026-10-19)
14. Gemini 3.8 Flash — Google. https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash (accessed 2026-10-19)
15. gpt-oss-120b — OpenAI. https://developers.openai.com/api/docs/models/gpt-oss-120b (accessed 2026-10-19)
16. GPT-Image-2.5 Sunburst — OpenAI. https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst (accessed 2026-10-19)
17. Pricing — Black Forest Labs. https://docs.bfl.ai/quick_start/pricing (accessed 2026-10-19)
18. Deprecations — OpenAI. https://developers.openai.com/api/docs/deprecations (accessed 2026-10-19)
19. Gemini deprecations — Google. https://ai.google.dev/gemini-api/docs/deprecations (accessed 2026-10-19)
20. API pricing — ElevenLabs. https://elevenlabs.io/pricing/api (accessed 2026-10-19)